技术领域Technical Field
本发明涉及数据检测技术领域，具体而言，涉及一种检测装置、检测方法、鞋垫、训练方法和识别方法。The present invention relates to the technical field of data detection, and in particular to a detection device, a detection method, an insole, a training method and an identification method.
背景技术Background Art
压电薄膜传感器是一种动态应变传感器，用于将受到的压力转换为电信号。现有的压电薄膜传感器仅能检测是否存在压力刺激，以及压力的大小，无法对受到的外部刺激做进一步的分析，检测能力较差，功能比较单一。Piezoelectric film sensors are a type of dynamic strain sensor that converts pressure into electrical signals. Existing piezoelectric film sensors can only detect whether there is pressure stimulation and the magnitude of the pressure, but cannot further analyze the external stimulation. They have poor detection capabilities and relatively simple functions.
发明内容Summary of the invention
本发明解决的问题是如何提高对外部刺激的检测能力。The problem solved by the present invention is how to improve the detection capability of external stimuli.
为解决上述问题，本发明提供一种检测装置、检测方法、鞋垫、训练方法和识别方法。In order to solve the above problems, the present invention provides a detection device, a detection method, an insole, a training method and an identification method.
第一方面，本发明提供了一种压电信号检测装置，包括相对设置的两层保护层、至少两层感觉层和绝缘层，各个所述感觉层设置在所述两层保护层之间，且每相邻的两层所述感觉层之间设置有一层所述绝缘层；In a first aspect, the present invention provides a piezoelectric signal detection device, comprising two protective layers, at least two sensing layers and an insulating layer arranged opposite to each other, each of the sensing layers being arranged between the two protective layers, and one insulating layer being arranged between each two adjacent sensing layers;
每层所述感觉层包括多个用于将外部刺激转换为电信号的感知单元，多个所述感知单元呈阵列分布，且每行或每列所述感知单元串联，串联的每行或每列所述感知单元的一端接地，另一端用于输出电信号，相邻的两层所述感觉层上的所述感知单元的串联方向之间具有标定夹角。Each sensory layer includes a plurality of sensing units for converting external stimuli into electrical signals. The plurality of sensing units are distributed in an array, and the sensing units in each row or column are connected in series. One end of the sensing units in each row or column of the series connection is grounded, and the other end is used to output electrical signals. There is a calibrated angle between the series connection directions of the sensing units on two adjacent sensory layers.
可选地，所述保护层和所述绝缘层包括聚二甲基硅氧烷，所述感知单元包括六氟丙烯与氧化石墨烯掺杂的聚偏氟乙烯。Optionally, the protective layer and the insulating layer include polydimethylsiloxane, and the sensing unit includes polyvinylidene fluoride doped with hexafluoropropylene and graphene oxide.
第二方面，本发明提供了一种外部刺激检测方法，基于如上所述的压电信号检测装置，包括：In a second aspect, the present invention provides an external stimulus detection method, based on the piezoelectric signal detection device as described above, comprising:
获取所述压电信号检测装置中各个感觉层在外部刺激作用下的输出电压；Obtaining the output voltage of each sensory layer in the piezoelectric signal detection device under external stimulation;
将上层感觉层的输出电压与第一预设阈值进行对比；comparing the output voltage of the upper sensory layer with a first preset threshold;
当所述上层感觉层的输出电压大于或等于所述第一预设阈值时，若任一下层感觉层的输出电压小于第二预设阈值，则表示所述外部刺激的类型为轻微滑动，其中所述第二预设阈值小于所述第一预设阈值，所述上层感觉层为最接近所述外部刺激作用部位的感觉层，所述下层感觉层为除所述上层感觉层以外的感觉层；When the output voltage of the upper sensory layer is greater than or equal to the first preset threshold, if the output voltage of any lower sensory layer is less than the second preset threshold, it indicates that the type of the external stimulus is slight sliding, wherein the second preset threshold is less than the first preset threshold, the upper sensory layer is the sensory layer closest to the site of action of the external stimulus, and the lower sensory layer is the sensory layer other than the upper sensory layer;
若所述上层感觉层的输出电压和任一所述下层感觉层的输出电压的符号相同，则所述外部刺激的类型为弯曲；If the output voltage of the upper sensory layer and the output voltage of any of the lower sensory layers have the same sign, then the type of the external stimulus is bending;
若所述上层感觉层的输出电压和任一所述下层感觉层的输出电压的符号相反，则表示所述外部刺激的类型为按压。If the output voltage of the upper sensory layer and the output voltage of any of the lower sensory layers have opposite signs, it indicates that the type of the external stimulus is pressing.
第三方面，本发明提供了一种鞋垫，所述鞋垫上设置有多个如上所述的压电信号检测装置。In a third aspect, the present invention provides an insole on which a plurality of piezoelectric signal detection devices as described above are arranged.
第四方面，本发明提供了一种分类器训练方法，基于如上所述的鞋垫，包括：In a fourth aspect, the present invention provides a classifier training method based on the insole as described above, comprising:
获取基于所述鞋垫执行不同的标定动作时，所述鞋垫上各个压电信号检测装置输出的传感器数据流；Acquire sensor data streams output by each piezoelectric signal detection device on the insole when performing different calibration actions based on the insole;
对各个的所述传感器数据流进行数据分割，获得多个传感器数据片段；Performing data segmentation on each of the sensor data streams to obtain a plurality of sensor data segments;
将各个所述传感器数据片段输入分类器，基于如上所述的外部刺激检测方法，对所述传感器数据片段进行特征提取，并采用提取得到的特征数据训练分类器，获得训练好的分类器。Each of the sensor data segments is input into a classifier, and based on the external stimulus detection method as described above, features of the sensor data segments are extracted, and the classifier is trained using the extracted feature data to obtain a trained classifier.
可选地，所述对所述传感器数据片段进行特征提取，并采用提取得到的特征数据训练分类器包括：Optionally, the extracting features from the sensor data segments and training a classifier using the extracted feature data includes:
前向传播步骤，提取各个所述传感器数据片段中与外部刺激对应的特征数据，根据所述特征数据确定所述标定动作为各个步态模板的概率，并确定概率最大的所述步态模板为预测步态；A forward propagation step, extracting feature data corresponding to the external stimulus in each of the sensor data segments, determining the probability that the calibration action is each gait template according to the feature data, and determining the gait template with the greatest probability as the predicted gait;
反向传播步骤，根据所述标定动作和所述预测步态做交叉熵损失，并根据所述交叉熵损失优化所述分类器；A back propagation step, performing a cross entropy loss according to the calibration action and the predicted gait, and optimizing the classifier according to the cross entropy loss;
循环重复所述前向传播步骤和所述反向传播步骤，直至所述损失不再下降，获得所述训练好的分类器。The forward propagation step and the back propagation step are repeated cyclically until the loss no longer decreases, thereby obtaining the trained classifier.
可选地，所述分类器包括依次连接的两个一维卷积层、最大池化层、展平层、LSTM层、挤压-激励计算单元、Softmax层和输出层，所述提取各个所述传感器数据片段中与外部刺激对应的特征数据，根据所述特征数据确定所述标定动作为各个步态模板的概率包括：Optionally, the classifier includes two one-dimensional convolutional layers, a maximum pooling layer, a flattening layer, an LSTM layer, a squeeze-excitation calculation unit, a Softmax layer and an output layer connected in sequence, and the extracting feature data corresponding to the external stimulus in each of the sensor data segments, and determining the probability that the calibration action is each gait template according to the feature data includes:
将多通道的所述传感器数据片段输入第一个所述一维卷积层，通过两个所述一维卷积层进行特征提取，并将提取得到的所有数据组成特征图；Input the multi-channel sensor data fragments into the first one-dimensional convolution layer, perform feature extraction through two one-dimensional convolution layers, and form a feature map with all the extracted data;
将所述特征图输入所述最大池化层，对所述特征图的每个子区域进行特征提取，获得与外部刺激对应的所述特征数据；Inputting the feature map into the maximum pooling layer, performing feature extraction on each sub-region of the feature map, and obtaining the feature data corresponding to the external stimulus;
将所述特征数据输入所述展平层，通过所述展平层将所述特征数据整形成一维向量，将所述一维向量输入所述LSTM层进行处理，所述LSTM层包括多个LSTM单元；Input the feature data into the flattening layer, shape the feature data into a one-dimensional vector through the flattening layer, input the one-dimensional vector into the LSTM layer for processing, and the LSTM layer includes a plurality of LSTM units;
将各个所述LSTM单元输出的数据输入挤压-激励计算单元进行加权，将加权后的数据输入所述Softmax层，确定所述标定动作为各个所述步态模板的概率；Input the data output by each LSTM unit into the squeeze-excitation calculation unit for weighting, input the weighted data into the Softmax layer, and determine the probability that the calibration action is each gait template;
所述输出层输出概率最大的所述步态模板，概率最大的所述步态模板为所述预测步态。The output layer outputs the gait template with the highest probability, and the gait template with the highest probability is the predicted gait.
可选地，所述挤压-激励计算单元包括依次连接的第一全连接层、ReLU层、第二全连接层和Sigmoid层。Optionally, the squeeze-excitation calculation unit includes a first fully connected layer, a ReLU layer, a second fully connected layer and a Sigmoid layer connected in sequence.
可选地，所述分类器还包括多个脱落层，其中，两个所述脱落层设置在第二个所述一维卷积层和所述最大池化层之间，两个所述脱落层设置在所述LSTM层和第一个所述全连接层之间。Optionally, the classifier further includes a plurality of dropout layers, wherein two of the dropout layers are arranged between the second one-dimensional convolutional layer and the maximum pooling layer, and two of the dropout layers are arranged between the LSTM layer and the first fully connected layer.
第五方面，本发明提供了一种步态识别方法，包括：In a fifth aspect, the present invention provides a gait recognition method, comprising:
获取用户完成当前动作时鞋垫上各个压电信号采集装置采集的传感器数据；Acquire sensor data collected by each piezoelectric signal collection device on the insole when the user completes the current action;
将所有所述传感器数据输入训练好的分类器，确定所述当前动作对应的步态，其中，所述训练好的分类器采用如上所述的分类器训练方法训练得到。All the sensor data are input into a trained classifier to determine the gait corresponding to the current action, wherein the trained classifier is trained using the classifier training method described above.
本发明的检测装置、检测方法、鞋垫、训练方法和识别方法的有益效果是：感觉层设置在两层保护层之间，可以保护感觉层上的感知单元，每相邻的两层感觉层之间设置有绝缘层，绝缘层可用作中性层，且能够避免感觉层之间发生串扰。感觉层中的每行或每列感知单元串联，能够简化电路连接结构，相邻两层感觉层的感知单元串联方向具有标定夹角，使得在同一外部刺激下，不同的感觉层输出的电信号不同，可根据各层感觉层输出的电信号的大小和符号判断外部刺激的类型，有利于对外部刺激做进一步的分析，提高了对外部刺激的检测能力。The detection device, detection method, insole, training method and identification method of the present invention have the following beneficial effects: the sensory layer is arranged between two protective layers, which can protect the sensory units on the sensory layer; an insulating layer is arranged between each two adjacent sensory layers, which can be used as a neutral layer and can avoid crosstalk between the sensory layers. The sensory units in each row or column in the sensory layer are connected in series, which can simplify the circuit connection structure; the series connection direction of the sensory units of two adjacent sensory layers has a calibrated angle, so that under the same external stimulus, different sensory layers output different electrical signals, and the type of external stimulus can be judged according to the size and sign of the electrical signals output by each sensory layer, which is conducive to further analysis of the external stimulus and improves the detection capability of the external stimulus.
附图说明BRIEF DESCRIPTION OF THE DRAWINGS
图1为本发明实施例的一种压电信号检测装置的结构示意图；FIG1 is a schematic structural diagram of a piezoelectric signal detection device according to an embodiment of the present invention;
图2为本发明实施例的上层感觉层和下层感觉层的结构示意图；FIG2 is a schematic diagram of the structure of the upper sensory layer and the lower sensory layer according to an embodiment of the present invention;
图3为本发明另一实施例的一种外部刺激检测方法的流程示意图；FIG3 is a schematic flow chart of an external stimulus detection method according to another embodiment of the present invention;
图4为本发明实施例的压电信号检测装置在压力作用下的结构示意图；FIG4 is a schematic structural diagram of a piezoelectric signal detection device under pressure according to an embodiment of the present invention;
图5为本发明实施例的轻微滑动刺激下两层感应层的响应曲线示意图；FIG5 is a schematic diagram of response curves of two sensing layers under slight sliding stimulation according to an embodiment of the present invention;
图6为本发明实施例的按压刺激下两层感应层的响应曲线示意图；FIG6 is a schematic diagram of response curves of two sensing layers under pressure stimulation according to an embodiment of the present invention;
图7为本发明实施例的压电信号检测装置在弯曲作用下的结构示意图；7 is a schematic structural diagram of a piezoelectric signal detection device under bending according to an embodiment of the present invention;
图8为本发明实施例的弯曲刺激下两层感应层的响应曲线示意图；FIG8 is a schematic diagram of response curves of two sensing layers under bending stimulation according to an embodiment of the present invention;
图9为本发明实施例的弯曲半径和感应层输出电压之间的关系示意图；FIG9 is a schematic diagram showing the relationship between the bending radius and the output voltage of the sensing layer according to an embodiment of the present invention;
图10为本发明又一实施例的一种鞋垫的结构示意图；FIG10 is a schematic structural diagram of an insole according to another embodiment of the present invention;
图11为本发明又一实施例的一种分类器训练方法的流程示意图；FIG11 is a schematic flow chart of a classifier training method according to another embodiment of the present invention;
图12为本发明又一实施例的步态识别方法的流程示意图。FIG. 12 is a flow chart of a gait recognition method according to another embodiment of the present invention.
具体实施方式DETAILED DESCRIPTION
为使本发明的上述目的、特征和优点能够更为明显易懂，下面结合附图对本发明的具体实施例做详细的说明。In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.
需要说明的是，本发明的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本发明的实施例能够以除了在这里图示或描述的那些以外的顺序实施。It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein.
矫形鞋垫是改变地面来适应穿着者的个人足部形状，用于矫正用户的步态等。为了达到更好的矫正效果，可在矫形鞋垫上设置压电薄膜传感器，以实时监控用户步态，在步态不正确时推送消息至终端，提醒用户矫正步态。这种方法可应用在多种领域中，例如患者的康复训练，运动过程监测和体感游戏等。压电薄膜传感器是一种动态应变传感器，现有的压电薄膜传感器仅能检测是否存在压力刺激，功能比较单一，不利于对受到的外部刺激做进一步的分析，进而不利于分析不同的动作和步态之间的关系，以及对步态的影响。Orthopedic insoles change the ground to adapt to the wearer's individual foot shape, and are used to correct the user's gait, etc. In order to achieve a better correction effect, a piezoelectric film sensor can be set on the orthopedic insole to monitor the user's gait in real time, and push messages to the terminal when the gait is incorrect to remind the user to correct the gait. This method can be applied in a variety of fields, such as patient rehabilitation training, exercise process monitoring, and somatosensory games. Piezoelectric film sensors are a kind of dynamic strain sensors. Existing piezoelectric film sensors can only detect whether there is pressure stimulation, and the function is relatively simple, which is not conducive to further analysis of the external stimulation received, and thus is not conducive to analyzing the relationship between different actions and gaits, as well as the impact on gait.
如图1和图2所示，本发明实施例提供的一种压电信号检测装置，包括相对设置的两层保护层、至少两层感觉层和绝缘层，所述感觉层设置在所述两层保护层之间，且每相邻的两层所述感觉层之间设置有一层所述绝缘层；As shown in FIG. 1 and FIG. 2 , a piezoelectric signal detection device provided by an embodiment of the present invention comprises two protective layers, at least two sensing layers and an insulating layer arranged opposite to each other, wherein the sensing layer is arranged between the two protective layers, and an insulating layer is arranged between each two adjacent sensing layers;
每层所述感觉层包括多个用于将外部刺激转换为电信号的感知单元，多个所述感知单元呈阵列分布，且每行或每列所述感知单元串联，串联的每行或每列所述感知单元的一端接地，另一端用于输出电信号，相邻的两层所述感觉层上的所述感知单元的串联方向之间具有标定夹角。Each sensory layer includes a plurality of sensing units for converting external stimuli into electrical signals. The plurality of sensing units are distributed in an array, and the sensing units in each row or column are connected in series. One end of the sensing units in each row or column of the series connection is grounded, and the other end is used to output electrical signals. There is a calibrated angle between the series connection directions of the sensing units on two adjacent sensory layers.
本实施例中，感觉层设置在两层保护层之间，可以保护感觉层上的感知单元，每相邻的两层感觉层之间设置有绝缘层，绝缘层可用作中性层，且能够避免感觉层之间发生串扰。感觉层中的每行或每列感知单元串联，能够简化电路连接结构，相邻两层感觉层的感知单元串联方向具有标定夹角，使得在同一外部刺激下，不同的感觉层输出的电信号不同，可根据各层感觉层输出的电信号的大小和符号判断外部刺激的类型，有利于对外部刺激做进一步的分析，提高了对外部刺激的检测能力。In this embodiment, the sensory layer is arranged between two protective layers, which can protect the sensory units on the sensory layer. An insulating layer is arranged between each two adjacent sensory layers, which can be used as a neutral layer and can avoid crosstalk between the sensory layers. Each row or column of sensory units in the sensory layer is connected in series, which can simplify the circuit connection structure. The series connection direction of the sensory units of two adjacent sensory layers has a calibrated angle, so that under the same external stimulus, different sensory layers output different electrical signals. The type of external stimulus can be judged according to the size and sign of the electrical signals output by each sensory layer, which is conducive to further analysis of the external stimulus and improves the detection capability of the external stimulus.
具体地，保护层用于保护内部的感觉层和传递外部刺激，厚度可优选为100μm，弹性模量可优选为2.6MPa，感觉层的厚度可优选为30μm，弹性模量可优选为26-33MPa，绝缘层的厚度可优选为100至300μm，绝缘层一方面用于避免两层感觉层之间发生串扰，另一方面相对更厚的绝缘层保证了应力中性层远离两层感觉层，使每一层感觉层在弯曲刺激下只能拉伸或压缩，进而提高输出电压。Specifically, the protective layer is used to protect the internal sensory layer and transmit external stimuli. The thickness may preferably be 100 μm and the elastic modulus may preferably be 2.6 MPa. The thickness of the sensory layer may preferably be 30 μm and the elastic modulus may preferably be 26-33 MPa. The thickness of the insulating layer may preferably be 100 to 300 μm. On the one hand, the insulating layer is used to avoid crosstalk between the two sensory layers. On the other hand, the relatively thicker insulating layer ensures that the stress-neutral layer is away from the two sensory layers, so that each sensory layer can only be stretched or compressed under bending stimulation, thereby increasing the output voltage.
如图2所示，以两层感觉层为例，每层感觉层中，每行或每列感知单元串联，形成一个通道，且两层感觉层上感知单元的串联方向具有一定的夹角，可优选为90度，即两层感觉层上感知单元的串联方向正交。As shown in Figure 2, taking two sensory layers as an example, in each sensory layer, each row or column of sensory units are connected in series to form a channel, and the series connection directions of the sensory units on the two sensory layers have a certain angle, which can be preferably 90 degrees, that is, the series connection directions of the sensory units on the two sensory layers are orthogonal.
可将每个通道一端的感知单元串联，然后接地，每个通道的另一端作为该通道感知单元的输出，每个通道的输出不相连，避免发生串扰。这种连接方式一方面避免了受到外部刺激时多个通道内部的局部变形，使应力传递更加均匀，另一方面，相较于将各个感知单元分别作为一个输出通道，能够减少导线数量，简化电路结构，节省成本。The sensing units at one end of each channel can be connected in series and then grounded. The other end of each channel serves as the output of the sensing unit of the channel. The outputs of each channel are not connected to avoid crosstalk. On the one hand, this connection method avoids local deformation inside multiple channels when subjected to external stimuli, making stress transmission more uniform. On the other hand, compared with using each sensing unit as an output channel, it can reduce the number of wires, simplify the circuit structure, and save costs.
可选地，所述保护层和所述绝缘层包括聚二甲基硅氧烷(PDMS)，所述感知单元包括六氟丙烯与氧化石墨烯掺杂的聚偏氟乙烯(PVDF-HFP/GO)。Optionally, the protective layer and the insulating layer include polydimethylsiloxane (PDMS), and the sensing unit includes polyvinylidene fluoride doped with hexafluoropropylene and graphene oxide (PVDF-HFP/GO).
具体地，聚二甲基硅氧烷无毒，具有良好的生物相容性，且成本较低，能够模拟皮肤组织和真皮层。聚偏氟乙烯具有超透明、高柔韧性和无铅生物相容性等优点。Specifically, polydimethylsiloxane is non-toxic, has good biocompatibility, is low-cost, and can simulate skin tissue and dermis. Polyvinylidene fluoride has the advantages of ultra-transparency, high flexibility, and lead-free biocompatibility.
本压电信号检测装置不仅可以识别针对单个感知单元的刺激，还可以针对多个感知单元的刺激。首先，对每个通道进行扫描，并判断上层感觉层的通道和下层感觉层的同时是否都被激活。根据经验，阈值可设置为0.4V，输出电压高于阈值的信号被认为是激活。若上、下感觉层同时激活，则多个感知单元之间的输出电压计算公式为V＝(Vt+Vb)/2，否则为V＝min(Vt,Vb)，其中，V为外界刺激引起的压电信号检测装置的电压输出，Vt为上层感觉层的输出电压，Vb为下层感觉层的输出电压。这里使用的感觉层弹性模量为600-700MPa，相比于常见的PVDF原膜(一般弹性模量为3GPa左右)具有更好的动态拉伸能力，可以通过六氟丙烯与氧化石墨烯复合材料成膜。This piezoelectric signal detection device can not only identify stimulation for a single sensory unit, but also for multiple sensory units. First, scan each channel and determine whether the channels of the upper sensory layer and the lower sensory layer are activated at the same time. According to experience, the threshold can be set to 0.4V, and the signal with an output voltage higher than the threshold is considered to be activated. If the upper and lower sensory layers are activated at the same time, the output voltage calculation formula between multiple sensory units is V = (Vt + Vb) / 2, otherwise it is V = min (Vt, Vb), where V is the voltage output of the piezoelectric signal detection device caused by external stimulation, Vt is the output voltage of the upper sensory layer, and Vb is the output voltage of the lower sensory layer. The elastic modulus of the sensory layer used here is 600-700MPa, which has better dynamic tensile ability than the common PVDF original film (generally with an elastic modulus of about 3GPa), and can be formed into a film through a composite material of hexafluoropropylene and graphene oxide.
如图3所示，本发明另一实施例提供的一种外部刺激检测方法，基于如上所述的压电信号检测装置，包括：As shown in FIG3 , another embodiment of the present invention provides an external stimulus detection method, based on the piezoelectric signal detection device as described above, comprising:
步骤S110，获取所述压电信号检测装置中各个感觉层在外部刺激作用下的输出电压；Step S110, obtaining the output voltage of each sensory layer in the piezoelectric signal detection device under external stimulation;
步骤S120，将上层感觉层的输出电压与第一预设阈值进行对比；Step S120, comparing the output voltage of the upper sensory layer with a first preset threshold;
步骤S130，当所述上层感觉层的输出电压大于或等于所述第一预设阈值时，若任一所述下层感觉层的输出电压小于第二预设阈值，则表示所述外部刺激的类型为轻微滑动，其中所述第二预设阈值小于所述第一预设阈值，所述上层感觉层为最接近所述外部刺激作用部位的感觉层，所述下层感觉层为除所述上层感觉层以外的感觉层；Step S130, when the output voltage of the upper sensory layer is greater than or equal to the first preset threshold, if the output voltage of any of the lower sensory layers is less than the second preset threshold, it indicates that the type of the external stimulus is slight sliding, wherein the second preset threshold is less than the first preset threshold, the upper sensory layer is the sensory layer closest to the site of action of the external stimulus, and the lower sensory layer is the sensory layer other than the upper sensory layer;
若所述上层感觉层的输出电压和所述下层感觉层的输出电压的符号相同，则所述外部刺激的类型为弯曲；If the output voltage of the upper sensory layer and the output voltage of the lower sensory layer have the same sign, then the type of the external stimulus is bending;
若所述上层感觉层的输出电压和所述下层感觉层的输出电压的符号相反，则表示所述外部刺激的类型为按压。If the output voltage of the upper sensory layer and the output voltage of the lower sensory layer have opposite signs, it indicates that the type of the external stimulation is pressing.
具体地，以两层感觉层为例，压电信号检测装置检测的外部刺激包括轻微滑动、按压和弯曲。如图4和图5所示，当压电信号检测装置受到轻微滑动刺激时，作用在压电信号检测装置上的压力很小，上层感觉层有较小的电压输出，下层感觉层几乎没有电压输出。如图4和图6所示，当压电信号检测装置受到按压刺激时，两层感觉层均处于压应力状态，最大应力位于按压刺激作用区域处的感知单元，同时由于绝缘层的缓冲作用，上层感觉层的感知单元的应力会大于下层感觉层感知单元的应力，即上层感觉层的输出电压会更大，同时从应力状态、压电本构关系、极化方向和电路连接可以看出，两层感觉层的输出电压符号相反(相位差为180°)。如图7和图8所示，当压电信号检测装置受到弯曲刺激时，中性层位于两层感觉层之间的绝缘层，导致两个感觉层的应力状态相反，一个为张力，一个为压缩力。同时感觉层产生的输出电压可反应弯曲方向和弯曲半径，如图中所示，弯曲方向的角度和弯曲半径可通过感觉层输出电压最大值来确定，当上层感觉层与下层感觉层输出电压一致时，即可反推弯曲方向的角度和弯曲半径。如图9所示为弯曲角度为10°时，不同弯曲半径下两层感应层的输出电压曲线，随着弯曲半径加大，上层感觉层的压力降低，张力增大，而下层感觉层的压力增大，输出电压增大。当弯曲角度为45°时，上下两层感觉层应力接近，此时两层输出层的输出电压的幅值也接近。这是对于多层感觉层，由于最上层感觉层在0°时有效弯曲，而最有效的弯曲方向时最下层感觉层在90°处弯曲，在45°时，感觉层结构及其施加的荷载近似对称，因此两层的应力分布此时较为接近。考虑应力状态、压电本构关系、极化方向和电路连接，在弯曲刺激下，两层感觉层的输出电压是同相同步的，输出电压同时取决于弯曲角度和弯曲半径。Specifically, taking two sensory layers as an example, the external stimuli detected by the piezoelectric signal detection device include slight sliding, pressing and bending. As shown in Figures 4 and 5, when the piezoelectric signal detection device is stimulated by slight sliding, the pressure acting on the piezoelectric signal detection device is very small, the upper sensory layer has a small voltage output, and the lower sensory layer has almost no voltage output. As shown in Figures 4 and 6, when the piezoelectric signal detection device is stimulated by pressing, both sensory layers are in a compressive stress state, and the maximum stress is located in the sensing unit at the area of the pressing stimulation. At the same time, due to the buffering effect of the insulating layer, the stress of the sensing unit of the upper sensory layer will be greater than the stress of the sensing unit of the lower sensory layer, that is, the output voltage of the upper sensory layer will be greater. At the same time, it can be seen from the stress state, piezoelectric constitutive relationship, polarization direction and circuit connection that the output voltage signs of the two sensory layers are opposite (the phase difference is 180°). As shown in Figures 7 and 8, when the piezoelectric signal detection device is stimulated by bending, the neutral layer is located in the insulating layer between the two sensory layers, resulting in opposite stress states of the two sensory layers, one is tension and the other is compression. At the same time, the output voltage generated by the sensory layer can reflect the bending direction and bending radius. As shown in the figure, the angle of the bending direction and the bending radius can be determined by the maximum value of the output voltage of the sensory layer. When the output voltages of the upper sensory layer and the lower sensory layer are consistent, the angle of the bending direction and the bending radius can be reversed. As shown in Figure 9, the output voltage curves of the two sensing layers under different bending radii when the bending angle is 10°. As the bending radius increases, the pressure of the upper sensory layer decreases, the tension increases, and the pressure of the lower sensory layer increases, and the output voltage increases. When the bending angle is 45°, the stresses of the upper and lower sensory layers are close, and the amplitudes of the output voltages of the two output layers are also close. This is for multiple sensory layers. Since the uppermost sensory layer is effectively bent at 0°, and the most effective bending direction is when the lowermost sensory layer is bent at 90°, at 45°, the sensory layer structure and the load applied to it are approximately symmetrical, so the stress distribution of the two layers is relatively close at this time. Considering the stress state, piezoelectric constitutive relationship, polarization direction and circuit connection, under bending stimulation, the output voltages of the two sensory layers are in phase and synchronized, and the output voltage depends on both the bending angle and the bending radius.
可根据实际应用需求，设计保护层和绝缘层的厚度和杨氏模量，在相同的弯曲方向和弯曲半径下，由于较厚的绝缘层增加了感知单元的刚度，保护层较厚的应力略高。The thickness and Young's modulus of the protective layer and the insulating layer can be designed according to actual application requirements. Under the same bending direction and bending radius, the stress of the thicker protective layer is slightly higher because the thicker insulating layer increases the stiffness of the sensing unit.
综上，在轻微滑动刺激下，上层感觉层有幅值较小的电压输出，而下层感觉层输出电压接近于零；在按压刺激下，上、下感觉层输出电压波形为180°相移；在弯曲刺激下，上、下感觉层输出电压相移相同，当弯曲角度达到45°时，上下两层感觉层输出电压同相等大；其它角度同相，但大小不同。In summary, under slight sliding stimulation, the upper sensory layer has a voltage output with a smaller amplitude, while the output voltage of the lower sensory layer is close to zero; under pressing stimulation, the output voltage waveforms of the upper and lower sensory layers are 180° phase-shifted; under bending stimulation, the output voltages of the upper and lower sensory layers have the same phase shift, and when the bending angle reaches 45°, the output voltages of the upper and lower sensory layers are in phase and equal in magnitude; other angles are in phase but different in magnitude.
首先将上层感觉层的输出电压和第一预设阈值进行对比，第一预设阈值可优选为0.25V，当上层感觉层的输出电压小于第一预设阈值时，说明外部刺激过小；当上层感觉层的输出电压大于或等于第一预设阈值时，说明上层感觉层有通道被激活，此时将下层感觉层的输出电压与第二预设阈值进行对比，第二预设阈值可优选为0.2V，若下层感觉层的输出电压小于第二预设阈值，表示外部刺激为轻微滑动；若下层感觉层的输出电压大于或等于第二预设阈值，则判断上层感觉层的输出电压和下层感觉层的输出电压的符号是否相同，若是，则外部刺激为弯曲，若否，则外部刺激为按压。First, the output voltage of the upper sensory layer is compared with the first preset threshold value, which may preferably be 0.25V. When the output voltage of the upper sensory layer is less than the first preset threshold value, it indicates that the external stimulation is too small. When the output voltage of the upper sensory layer is greater than or equal to the first preset threshold value, it indicates that a channel in the upper sensory layer is activated. At this time, the output voltage of the lower sensory layer is compared with the second preset threshold value, which may preferably be 0.2V. If the output voltage of the lower sensory layer is less than the second preset threshold value, it indicates that the external stimulation is a slight sliding. If the output voltage of the lower sensory layer is greater than or equal to the second preset threshold value, it is determined whether the signs of the output voltages of the upper sensory layer and the lower sensory layer are the same. If so, the external stimulation is bending. If not, the external stimulation is pressing.
本实施例中，结合各层感应层的输出电压进行分析，可迅速确定外部刺激的类型，实现了对外部刺激的进一步分析，提高了对外部刺激的检测能力。In this embodiment, the type of external stimulus can be quickly determined by analyzing the output voltages of each sensing layer, thereby achieving further analysis of the external stimulus and improving the detection capability of the external stimulus.
如图10所示，本发明又一实施例提供的一种鞋垫上设置有多个如上所述的压电信号检测装置。As shown in FIG. 10 , another embodiment of the present invention provides an insole on which a plurality of piezoelectric signal detection devices as described above are arranged.
具体地，可根据实际应用需求，在鞋垫上布设压电信号检测装置，为了保证数据采集精度和数量，压电信号采集装置的数量建议不少于8个，同时考虑到成本和接线复杂度，数量建议不超过20个。Specifically, according to actual application requirements, a piezoelectric signal detection device can be arranged on the insole. In order to ensure the accuracy and quantity of data collection, the number of piezoelectric signal collection devices is recommended to be no less than 8. Taking into account the cost and wiring complexity, the number is recommended not to exceed 20.
本实施例中，在鞋垫上布设多个压电信号检测装置，通过压电信号检测装置检测测试者脚底各个位置的受力数据，可通过对这些受力数据的进一步分析，确定测试者的脚部运动情况，例如步态等，扩大了压电信号检测装置的应用范围。In this embodiment, multiple piezoelectric signal detection devices are arranged on the insole, and the piezoelectric signal detection devices are used to detect the force data of various positions on the sole of the tester's foot. Through further analysis of these force data, the tester's foot movement, such as gait, etc., can be determined, thereby expanding the application scope of the piezoelectric signal detection device.
如图11所示，本发明又一实施例提供的一种分类器训练方法，基于如上所述的鞋垫，包括：As shown in FIG. 11 , a classifier training method provided by another embodiment of the present invention is based on the insole as described above, comprising:
步骤S210，获取测试者执行不同的标定动作时，所述鞋垫上各个压电信号检测装置输出的传感器数据流。Step S210, obtaining the sensor data stream output by each piezoelectric signal detection device on the insole when the tester performs different calibration actions.
具体地，测试者穿着垫有上述鞋垫的鞋子，脚执行对应的标定动作。Specifically, the tester wears shoes with the above-mentioned insoles, and the feet perform corresponding calibration actions.
步骤S220，对各个的所述传感器数据流进行数据分割，获得多个传感器数据片段。Step S220: segment each of the sensor data streams to obtain a plurality of sensor data segments.
具体地，在进行数据分割前，可预先对各个压电信号检测装置输出的传感器数据流进行时间同步和滤波。数据分割是用于在传感器数据流中识别出包含了活动信息的数据段，可采用滑动窗口在传感器数据流上移动，截取多个传感器数据片段，可通过比较相邻两个传感器数据片段之间的差异，以找到对应活动信息的断点。或者还可以通过检测活动转换或活动边界，根据活动转换点或活动边界对传感器数据流进行数据分割，例如：压力刺激与压电信号检测装置的输出电压是线性关系，而弯曲刺激与输出电压的关系曲线是多项式曲线，由此可以判断用户运动过程中的状态转移点，进行数据分割。Specifically, before data segmentation, the sensor data streams output by each piezoelectric signal detection device can be time-synchronized and filtered in advance. Data segmentation is used to identify data segments containing activity information in the sensor data stream. A sliding window can be used to move on the sensor data stream to intercept multiple sensor data segments. The breakpoints corresponding to the activity information can be found by comparing the differences between two adjacent sensor data segments. Alternatively, the sensor data stream can be segmented according to the activity transition point or activity boundary by detecting activity transitions or activity boundaries. For example, the relationship between pressure stimulation and the output voltage of the piezoelectric signal detection device is linear, while the relationship curve between bending stimulation and the output voltage is a polynomial curve. This can determine the state transition point during the user's movement and perform data segmentation.
步骤S230，将各个所述传感器数据片段输入分类器，基于如上所述的外部刺激检测方法，对所述传感器数据片段进行特征提取，并采用提取得到的特征数据训练分类器，获得训练好的分类器。Step S230, inputting each of the sensor data segments into a classifier, performing feature extraction on the sensor data segments based on the external stimulus detection method as described above, and using the extracted feature data to train the classifier to obtain a trained classifier.
本实施例中，获取测试者完成不同步态模板对应的标定动作时输出的传感器数据流，对传感器数据流进行数据分割处理后，用于训练分类器，分类器可基于机器学习模型建立，训练所需的数据少，精度高。In this embodiment, the sensor data stream output when the tester completes the calibration actions corresponding to different gait templates is obtained, and the sensor data stream is segmented and processed before being used to train a classifier. The classifier can be established based on a machine learning model, and the training requires less data and has higher accuracy.
可选地，所述对所述传感器数据片段进行特征提取，并采用提取得到的特征数据训练分类器包括：Optionally, the extracting features from the sensor data segments and training a classifier using the extracted feature data includes:
前向传播步骤，提取各个所述传感器数据片段中与外部刺激对应的特征数据，根据所述特征数据确定所述标定动作为各个步态模板的概率，并确定概率最大的所述步态模板为预测步态。The forward propagation step extracts the feature data corresponding to the external stimulus in each of the sensor data segments, determines the probability that the calibration action is each gait template based on the feature data, and determines the gait template with the highest probability as the predicted gait.
具体地，基于上述外部刺激检测方法，识别轻微滑动刺激、按压刺激和弯曲刺激，在传感器数据片段中提取出与外部刺激对应的特征数据。Specifically, based on the above external stimulus detection method, slight sliding stimulus, pressing stimulus and bending stimulus are identified, and feature data corresponding to the external stimulus are extracted from the sensor data segment.
反向传播步骤，根据所述标定动作和所述预测步态做交叉熵损失，并根据所述交叉熵损失优化所述分类器；A back propagation step, performing a cross entropy loss according to the calibration action and the predicted gait, and optimizing the classifier according to the cross entropy loss;
循环重复所述前向传播步骤和所述反向传播步骤，直至所述损失不再下降，获得所述训练好的分类器。The forward propagation step and the back propagation step are repeated cyclically until the loss no longer decreases, thereby obtaining the trained classifier.
可选地，所述分类器包括依次连接的两个一维卷积层、最大池化层、展平层、LSTM层、挤压-激励计算单元、Softmax层和输出层，所述提取各个所述传感器数据片段中与外部刺激对应的特征数据，根据所述特征数据确定所述标定动作为各个步态模板的概率包括：Optionally, the classifier includes two one-dimensional convolutional layers, a maximum pooling layer, a flattening layer, an LSTM layer, a squeeze-excitation calculation unit, a Softmax layer and an output layer connected in sequence, and the extracting feature data corresponding to the external stimulus in each of the sensor data segments, and determining the probability that the calibration action is each gait template according to the feature data includes:
将多通道的所述传感器数据片段输入第一个所述一维卷积层，通过两个所述一维卷积层进行特征提取，并将提取得到的所有数据组成特征图。The multi-channel sensor data segments are input into the first one-dimensional convolutional layer, feature extraction is performed through two one-dimensional convolutional layers, and all the extracted data are combined into a feature map.
具体地，将传感器数据片段从一维时间序列数据整形为二维矩阵，以满足Tensorflow或Keras(机器学习框架)计算框架下一维卷积层的输入尺寸要求，其中一个维度是时间步长，另一个维度是每个时间步长上的特征。采用两个一维卷积层，能够提高提取的特征的鲁棒性。一维卷积层是CNN(Convolutional Neural Networks,卷积神经网络)的一个变种，专门用于处理序列和时间序列数据。在一维卷积层中，卷积滤波器仅沿数据的时间方向移动，因此，一维卷积层能够从固定长度段的数据中导出特征。当应用于识别步态时，CNN相对于其他模型具有两个优势，局部依赖性和尺度不变性，局部依赖性表示附近的信号可能是相关的，而尺度不变性表示不同步距或频率的尺度不变。Specifically, the sensor data segments are reshaped from one-dimensional time series data into two-dimensional matrices to meet the input size requirements of the next dimensional convolutional layer in the Tensorflow or Keras (machine learning framework) computing framework, where one dimension is the time step and the other dimension is the feature at each time step. Using two one-dimensional convolutional layers can improve the robustness of the extracted features. The one-dimensional convolutional layer is a variant of CNN (Convolutional Neural Networks) that is specifically designed to process sequence and time series data. In a one-dimensional convolutional layer, the convolution filter only moves along the time direction of the data, so the one-dimensional convolutional layer can derive features from fixed-length segments of data. When applied to gait recognition, CNN has two advantages over other models, local dependence and scale invariance, where local dependence indicates that nearby signals may be related, and scale invariance indicates that the scale of different step lengths or frequencies is invariant.
将所述特征图输入所述最大池化层，对所述特征图的每个子区域进行特征提取，获得与外部刺激对应的所述特征数据。The feature map is input into the maximum pooling layer, and features are extracted for each sub-region of the feature map to obtain the feature data corresponding to the external stimulus.
具体地，在完成卷积之后，应用最大池化层从一维卷积层输出的特征图中的每个区域提取最重要的特征，可以减少特征的数量，加快训练过程。Specifically, after completing the convolution, a maximum pooling layer is applied to extract the most important features from each region in the feature map output by the one-dimensional convolutional layer, which can reduce the number of features and speed up the training process.
将所述特征数据输入所述展平层，通过所述展平层将所述特征数据整形成一维向量，将所述一维向量输入所述LSTM层进行处理，所述LSTM层包括多个LSTM单元。The feature data is input into the flattening layer, the feature data is shaped into a one-dimensional vector through the flattening layer, and the one-dimensional vector is input into the LSTM layer for processing, wherein the LSTM layer includes a plurality of LSTM units.
具体地，由于经过一维卷积层和最大池化层处理后的输出是二维矩阵，而LSTM层的输入尺寸要求是一维向量，因此采用展平层将最大池化层输出的第二特征数据重新展平成一维向量，作为一维时间序列数据输入到LSTM层。展平层可被视为CNN和LSTM层之间的桥接层，用于将二维数据转换成一维数据，统一LSTM层而不丢失信息。Specifically, since the output after the one-dimensional convolution layer and the maximum pooling layer is a two-dimensional matrix, and the input size of the LSTM layer is required to be a one-dimensional vector, the flattening layer is used to flatten the second feature data output by the maximum pooling layer into a one-dimensional vector and input it into the LSTM layer as one-dimensional time series data. The flattening layer can be regarded as a bridge layer between the CNN and LSTM layers, which is used to convert two-dimensional data into one-dimensional data and unify the LSTM layer without losing information.
将各个所述LSTM单元输出的数据输入挤压-激励计算单元进行加权，将加权后的数据输入所述Softmax层，确定所述标定动作为各个所述步态模板的概率。The data output by each LSTM unit is input into the squeeze-excitation calculation unit for weighting, and the weighted data is input into the Softmax layer to determine the probability that the calibration action is each gait template.
可选地，所述挤压-激励计算单元包括依次连接的第一全连接层、ReLU层、第二全连接层和Sigmoid层。Optionally, the squeeze-excitation calculation unit includes a first fully connected layer, a ReLU layer, a second fully connected layer and a Sigmoid layer connected in sequence.
具体地，每个LSTM单元对应一个信道的输入，使用两层大小不同的全连接层，通过对多个信号的输入进行缩放来对不同信道分配不同的权重。在第一个全连接层后采用校正线性单元ReLu层激活，一方面，由于它的梯度是非饱和的，可以有效加快梯度下降的收敛速度，满足快速收敛的需求；另一方面，ReLU函数通过梯度为0或1来解决梯度消失的问题。当对不同信道分别加权后，ReLU层的输出作为第二个全连层的输入，再使用第二全连层和Sigmoid层激活，并通过SoftMax层对维度进行还原并输出分类标签。所有类别标签的概率之和为1，具有最高概率的类别标签将是模型的最终预测类别标签。Specifically, each LSTM unit corresponds to the input of a channel, and two fully connected layers of different sizes are used to assign different weights to different channels by scaling the input of multiple signals. After the first fully connected layer, the rectified linear unit ReLu layer is used for activation. On the one hand, since its gradient is non-saturated, it can effectively speed up the convergence speed of gradient descent and meet the requirements of fast convergence; on the other hand, the ReLU function solves the problem of gradient disappearance by using a gradient of 0 or 1. After weighting different channels separately, the output of the ReLU layer is used as the input of the second fully connected layer, and then activated by the second fully connected layer and the Sigmoid layer, and the dimension is restored through the SoftMax layer and the classification label is output. The sum of the probabilities of all category labels is 1, and the category label with the highest probability will be the final predicted category label of the model.
所述输出层输出概率最大的所述步态模板，概率最大的所述步态模板为所述预测步态。The output layer outputs the gait template with the highest probability, and the gait template with the highest probability is the predicted gait.
具体地，Softmax层常用在模型的最后一层，是分类问题中常用的激活函数，输出步态模板对应的动作为各个类别标签的概率，并且所有类别标签的概率之和为1，其中概率最高的类别标签就是模型的预测类别标签，即模型的预测步态。Specifically, the Softmax layer is often used as the last layer of the model. It is an activation function commonly used in classification problems. It outputs the probability that the action corresponding to the gait template is each category label, and the sum of the probabilities of all category labels is 1. The category label with the highest probability is the predicted category label of the model, that is, the predicted gait of the model.
可选地，所述分类器还包括多个脱落层，其中，两个所述脱落层设置在第二个所述一维卷积层和所述最大池化层之间，两个所述脱落层设置在所述LSTM层和第一个所述全连接层之间。Optionally, the classifier further includes a plurality of dropout layers, wherein two of the dropout layers are arranged between the second one-dimensional convolutional layer and the maximum pooling layer, and two of the dropout layers are arranged between the LSTM layer and the first fully connected layer.
具体地，用相对较少的数据集训练神经网络时会导致训练数据过拟合，这是因为模型会学习训练数据中的统计噪声，当训练模型在测试或新数据被评估时会表现出较差的性能。因此，为了防止过拟合并减少泛化误差，深度学习框架中引入脱落层，模型学习鲁棒特征，可将脱落层的脱落率设置为0.5，表示脱落层随机选择输入单元的50％设置为零。Specifically, training a neural network with a relatively small dataset will lead to overfitting of the training data, because the model will learn the statistical noise in the training data, and will show poor performance when the trained model is tested or evaluated on new data. Therefore, in order to prevent overfitting and reduce generalization error, a dropout layer is introduced in the deep learning framework, and the model learns robust features. The dropout rate of the dropout layer can be set to 0.5, which means that the dropout layer randomly selects 50% of the input units to be set to zero.
可通过注意力机制，根据各个通道的权重对压电信号检测装置的数量和空间布局进行优化，以降低压电信号检测装置的数量和布线复杂度。还可根据实际应用需求调整压电信号检测装置的采样率，例如由于行动不便的老人行走步频较低，即可适当降低采样率，以降低能耗。The number and spatial layout of piezoelectric signal detection devices can be optimized according to the weights of each channel through the attention mechanism to reduce the number of piezoelectric signal detection devices and wiring complexity. The sampling rate of the piezoelectric signal detection device can also be adjusted according to actual application requirements. For example, since the walking frequency of elderly people with limited mobility is low, the sampling rate can be appropriately reduced to reduce energy consumption.
可预先根据不同应用场景，例如老年人步态监测或运动员步态监测等，建立通用受试让人群数据集和跨受试人群数据集，数据集包括测试者执行标定动作时压电信号检测装置输出的传感器数据流，根据不同用户群体的年龄、性别、业余/专业等因素对数据集中的数据进行分类，这种方法包含了使用者自身数据，鲁棒性较弱，但模型训练效果更好。通用受试人群数据集用于训练通用步态分类器，根据该通用步态分类器可用于完成不同个体的分类任务。跨受试人群数据集时排除受试者之外的数据集，可用于训练群体姿势分类器，能够省去校准过程，针对自身数据集以外的跨受试人群数据集建立模型，对模型架构的测试更为严格。A general subject population dataset and a cross-subject population dataset can be established in advance according to different application scenarios, such as gait monitoring of the elderly or athletes. The dataset includes the sensor data stream output by the piezoelectric signal detection device when the tester performs the calibration action. The data in the dataset is classified according to factors such as age, gender, amateur/professional, etc. of different user groups. This method includes the user's own data and has weaker robustness, but the model training effect is better. The general subject population dataset is used to train a general gait classifier, which can be used to complete classification tasks for different individuals. The cross-subject population dataset excludes the dataset outside the subject and can be used to train the group posture classifier, which can save the calibration process, build a model for the cross-subject population dataset other than its own dataset, and test the model architecture more strictly.
如图12所示，本发明又一实施例提供的一种步态识别方法，包括：As shown in FIG12 , another embodiment of the present invention provides a gait recognition method, comprising:
步骤S310，获取用户完成当前动作时鞋垫上各个压电信号采集装置采集的传感器数据；Step S310, obtaining sensor data collected by each piezoelectric signal collection device on the insole when the user completes the current action;
步骤S320，将所有所述传感器数据输入训练好的分类器，确定所述当前动作对应的步态，其中，所述训练好的分类器采用如上所述的分类器训练方法训练得到。Step S320, inputting all the sensor data into a trained classifier to determine the gait corresponding to the current action, wherein the trained classifier is trained using the classifier training method described above.
本实施例中，将鞋垫上各个压电信号采集装置采集的传感器数据输入到训练好的分类器中，确定概率最大的预测步态为当前动作对应的步态，能够迅速识别测试者的当前动作，当前动作对应的步态不正确时，能够迅速反馈给用户，提醒用户及时调整步态。In this embodiment, the sensor data collected by each piezoelectric signal acquisition device on the insole is input into the trained classifier to determine the predicted gait with the highest probability as the gait corresponding to the current action. The current action of the tester can be quickly identified. When the gait corresponding to the current action is incorrect, it can be quickly fed back to the user to remind the user to adjust the gait in time.
本发明又一实施例提供的一种外部刺激检测装置，包括：Another embodiment of the present invention provides an external stimulus detection device, comprising:
第一获取模块，用于获取所述压电信号检测装置中各个感觉层在外部刺激作用下的输出电压；A first acquisition module, used to acquire the output voltage of each sensory layer in the piezoelectric signal detection device under external stimulation;
处理模块，用于将上层感觉层的输出电压与第一预设阈值进行对比；当所述上层感觉层的输出电压大于或等于所述第一预设阈值时，若任一所述下层感觉层的输出电压小于第二预设阈值，则表示所述外部刺激的类型为轻微滑动，其中所述第二预设阈值小于所述第一预设阈值，所述上层感觉层为最接近所述外部刺激作用部位的感觉层，所述下层感觉层为除所述上层感觉层以外的感觉层；若所述上层感觉层的输出电压和所述下层感觉层的输出电压的符号相同，则所述外部刺激的类型为弯曲；若所述上层感觉层的输出电压和所述下层感觉层的输出电压的符号相反，则表示所述外部刺激的类型为按压。A processing module is used to compare the output voltage of the upper sensory layer with a first preset threshold value; when the output voltage of the upper sensory layer is greater than or equal to the first preset threshold value, if the output voltage of any of the lower sensory layers is less than the second preset threshold value, it indicates that the type of the external stimulation is slight sliding, wherein the second preset threshold value is less than the first preset threshold value, the upper sensory layer is the sensory layer closest to the site of action of the external stimulation, and the lower sensory layer is the sensory layer other than the upper sensory layer; if the output voltage of the upper sensory layer and the output voltage of the lower sensory layer have the same sign, the type of the external stimulation is bending; if the output voltage of the upper sensory layer and the output voltage of the lower sensory layer have opposite signs, it indicates that the type of the external stimulation is pressing.
本发明又一实施例提供的一种分类器训练装置，基于如上所述的鞋垫，包括：A classifier training device provided by another embodiment of the present invention is based on the insole as described above, and includes:
第二获取模块，用于获取测试者执行不同的标定动作时，所述鞋垫上各个压电信号检测装置输出的传感器数据流；A second acquisition module is used to acquire sensor data streams output by each piezoelectric signal detection device on the insole when the tester performs different calibration actions;
分割模块，用于对各个通道的所述传感器数据流进行数据分割，获得多个传感器数据片段；A segmentation module, used for performing data segmentation on the sensor data stream of each channel to obtain a plurality of sensor data segments;
训练模块，用于将各个所述传感器数据片段输入分类器，基于如上所述的外部刺激检测方法，对所述传感器数据片段进行特征提取，并采用提取得到的特征数据训练分类器，获得训练好的分类器。The training module is used to input each of the sensor data segments into a classifier, extract features from the sensor data segments based on the external stimulus detection method as described above, and use the extracted feature data to train the classifier to obtain a trained classifier.
本发明又一实施例提供的一种步态识别方法，包括：Another embodiment of the present invention provides a gait recognition method, comprising:
第三获取模块，用于获取用户完成当前动作时鞋垫上各个压电信号采集装置采集的传感器数据；The third acquisition module is used to acquire sensor data collected by each piezoelectric signal acquisition device on the insole when the user completes the current action;
识别模块，用于将所有所述传感器数据输入训练好的分类器，确定所述当前动作对应的步态，其中，所述训练好的分类器采用如上所述的分类器训练方法训练得到。The recognition module is used to input all the sensor data into a trained classifier to determine the gait corresponding to the current action, wherein the trained classifier is trained using the classifier training method described above.
本发明又一实施例提供的一种电子设备，包括存储器和处理器；所述存储器，用于存储计算机程序；所述处理器，用于当执行所述计算机程序时，实现如上所述的外部刺激检测方法，或所述分类器训练方法，或所述步态识别方法。Another embodiment of the present invention provides an electronic device, including a memory and a processor; the memory is used to store a computer program; the processor is used to implement the external stimulus detection method, the classifier training method, or the gait recognition method as described above when executing the computer program.
本发明又一实施例提供的一种计算机可读存储介质上存储有计算机程序，当所述计算机程序被处理器执行时，实现如上所述的外部刺激检测方法，或所述分类器训练方法，或所述步态识别方法。Another embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the external stimulus detection method, the classifier training method, or the gait recognition method as described above is implemented.
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的程序可存储于一计算机可读取存储介质中，该程序在执行时，可包括如上述各方法的实施例的流程。其中，所述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)或随机存储记忆体(Random AccessMemory，RAM)等。在本申请中，所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本发明实施例方案的目的。另外，在本发明各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以是两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc. In the present application, the unit described as a separate component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment of the present invention. In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit.
虽然本发明公开披露如上，但本发明公开的保护范围并非仅限于此。本领域技术人员在不脱离本发明公开的精神和范围的前提下，可进行各种变更与修改，这些变更与修改均将落入本发明的保护范围。Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the protection scope of the present invention.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110935805.5ACN113647937B (en) | 2021-08-16 | 2021-08-16 | Detection device, detection method, insole, training method and identification method |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110935805.5ACN113647937B (en) | 2021-08-16 | 2021-08-16 | Detection device, detection method, insole, training method and identification method |
| Publication Number | Publication Date |
|---|---|
| CN113647937A CN113647937A (en) | 2021-11-16 |
| CN113647937Btrue CN113647937B (en) | 2024-09-20 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110935805.5AActiveCN113647937B (en) | 2021-08-16 | 2021-08-16 | Detection device, detection method, insole, training method and identification method |
| Country | Link |
|---|---|
| CN (1) | CN113647937B (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114459334B (en)* | 2022-01-28 | 2023-10-10 | 北京科技大学 | Composite sensor for tensile strain vector monitoring and preparation method thereof |
| CN115965994B (en)* | 2022-12-20 | 2025-07-29 | 中国科学院自动化研究所 | Gait quantification method and device based on low-resolution monocular camera |
| CN119880203A (en)* | 2024-12-27 | 2025-04-25 | 上海交通大学 | Wide-working-condition multi-dimensional-force-representation flexible film piezoelectric sensor and signal analysis method |
| CN120426856A (en)* | 2025-06-05 | 2025-08-05 | 中国计量大学 | Fruit tree growth state monitoring method and device based on flexible sensor |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110274713A (en)* | 2019-06-13 | 2019-09-24 | 东华大学 | A kind of fiber base form adaptive passive electronic skin and preparation method thereof |
| CN110414366A (en)* | 2019-07-04 | 2019-11-05 | 东南大学 | A piezoresistive array and pressure distribution matching method based on dynamic signals |
| CN112906673A (en)* | 2021-04-09 | 2021-06-04 | 河北工业大学 | Lower limb movement intention prediction method based on attention mechanism |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101520355B (en)* | 2009-04-02 | 2011-05-18 | 张硕 | Plantar pressure distribution signal acquisition system |
| CN201382838Y (en)* | 2009-04-02 | 2010-01-13 | 张硕 | Large-area pressure distribution signal acquisition system |
| WO2010119441A2 (en)* | 2009-04-13 | 2010-10-21 | Wellsense Technologies | System and method for preventing decubitus ulcers |
| WO2012056405A2 (en)* | 2010-10-28 | 2012-05-03 | Enhanced Surface Dynamics, Inc. | Pressure sensor assembly and associated method for preventing the development of pressure injuries |
| WO2013008187A1 (en)* | 2011-07-13 | 2013-01-17 | Enhanced Surface Dynamics, Inc. | Methods and systems for the manufacture and initiation of a pressure detection mat |
| JP6835354B2 (en)* | 2016-03-18 | 2021-02-24 | 国立大学法人神戸大学 | Pressure sensor |
| CN106074074B (en)* | 2016-06-02 | 2019-02-12 | 哈尔滨易奥秘科技发展有限公司 | A kind of eccentric fiber composite elastic body and the device for healing and training with the elastomer |
| US10386224B2 (en)* | 2016-10-25 | 2019-08-20 | Studio 1 Labs Inc. | Flexible conductive apparatus and systems for detecting pressure |
| EP3558114A2 (en)* | 2016-12-21 | 2019-10-30 | Elwha Llc | Monitoring body movement or condition according to motion regimen with conformal electronics |
| CA3073289A1 (en)* | 2017-08-22 | 2019-02-28 | Kinetyx Sciences Inc. | Method and system for activity classification |
| US10378975B1 (en)* | 2018-01-27 | 2019-08-13 | Nextiles Inc. | Systems, methods, and devices for static and dynamic body measurements |
| CN109349737A (en)* | 2018-11-15 | 2019-02-19 | 北京大学深圳研究生院 | An insole for foot pressure collection |
| CN110333008A (en)* | 2019-07-17 | 2019-10-15 | 复旦大学 | capacitive touch sensor |
| CN110720921A (en)* | 2019-10-23 | 2020-01-24 | 京东方科技集团股份有限公司 | Electronic equipment |
| CN111504521B (en)* | 2020-05-07 | 2021-09-03 | 腾讯科技（深圳）有限公司 | Flexible capacitor array, preparation method thereof and capacitor array detection system |
| CN111982379B (en)* | 2020-07-31 | 2022-08-16 | 鹏城实验室 | Continuous elastic touch sensor and detection method thereof |
| CN112244819A (en)* | 2020-11-10 | 2021-01-22 | 浙大宁波理工学院 | A system and method for identifying abnormal gait in children based on plantar pressure array detection |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110274713A (en)* | 2019-06-13 | 2019-09-24 | 东华大学 | A kind of fiber base form adaptive passive electronic skin and preparation method thereof |
| CN110414366A (en)* | 2019-07-04 | 2019-11-05 | 东南大学 | A piezoresistive array and pressure distribution matching method based on dynamic signals |
| CN112906673A (en)* | 2021-04-09 | 2021-06-04 | 河北工业大学 | Lower limb movement intention prediction method based on attention mechanism |
| Publication number | Publication date |
|---|---|
| CN113647937A (en) | 2021-11-16 |
| Publication | Publication Date | Title |
|---|---|---|
| CN113647937B (en) | Detection device, detection method, insole, training method and identification method | |
| Javeed et al. | Wearable sensors based exertion recognition using statistical features and random forest for physical healthcare monitoring | |
| DelPreto et al. | A wearable smart glove and its application of pose and gesture detection to sign language classification | |
| Kour et al. | Computer-vision based diagnosis of Parkinson’s disease via gait: A survey | |
| Cai et al. | Gait phases recognition based on lower limb sEMG signals using LDA-PSO-LSTM algorithm | |
| Houmanfar et al. | Movement analysis of rehabilitation exercises: Distance metrics for measuring patient progress | |
| Assaleh et al. | Low complexity classification system for glove-based arabic sign language recognition | |
| Guo et al. | MCDCD: Multi-source unsupervised domain adaptation for abnormal human gait detection | |
| Kawsar et al. | A novel activity detection system using plantar pressure sensors and smartphone | |
| US20230040650A1 (en) | Real-time, fine-resolution human intra-gait pattern recognition based on deep learning models | |
| US12213777B2 (en) | Method and system for monitoring movement | |
| CN117690583B (en) | Interactive management system and method for rehabilitation nursing based on Internet of Things | |
| KR102350593B1 (en) | Apparatus and method for classifying gait pattern based on multi modal sensor using deep learning ensemble | |
| WO2021142532A1 (en) | Activity recognition with deep embeddings | |
| CN113947119A (en) | A method for detecting human gait using plantar pressure signal | |
| Bobin et al. | SVM based approach for the assessment of elbow flexion with smart textile sensor | |
| Sardari et al. | View-invariant pose analysis for human movement assessment from rgb data | |
| CN108681412B (en) | A kind of emotion recognition device and method based on array tactile sensor | |
| KR20200144996A (en) | Motor performance coashing system based on muscle synergy library and calculation method for muscle synergy characteristic based on muscle synergy library | |
| KR102302234B1 (en) | Apparatus and method for identifying individuals by performing discriminant analysis for various detection information | |
| Parle et al. | YogaWise: Enhancing Yoga with Intelligent Real Time Tracking using TensorFlow MoveNet | |
| KR101875472B1 (en) | Finger language recognition system and method | |
| Jeong et al. | Physical workout classification using wrist accelerometer data by deep convolutional neural networks | |
| Kathirgamanathan et al. | Towards Globalised Models for Exercise Classification using Inertial Measurement Units | |
| Xie et al. | Lightweight Football Motion Recognition and Intensity Analysis Using Low‐Cost Wearable Sensors |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | Effective date of registration:20250722 Address after:150001 Heilongjiang Province Harbin City Nangang District Shunyi Street No.7 2nd floor 3S area Patentee after:Harbin qiaoran Technology Co.,Ltd. Country or region after:China Address before:315048 No. 66, Yishilu Road, Gaoxin District, Ningbo City, Zhejiang Province Room 3-111, Building 2, Entrepreneurship Building Patentee before:Ningbo rongbotong Electromechanical Technology Co.,Ltd. Country or region before:China | |
| TR01 | Transfer of patent right |