CN110782461A - Mask-RCNN-based infrared image segmentation method for electric power equipment - Google Patents
Mask-RCNN-based infrared image segmentation method for electric power equipmentDownload PDFInfo
- Publication number
- CN110782461A CN110782461ACN201911027493.7ACN201911027493ACN110782461ACN 110782461 ACN110782461 ACN 110782461ACN 201911027493 ACN201911027493 ACN 201911027493ACN 110782461 ACN110782461 ACN 110782461A
- Authority
- CN
- China
- Prior art keywords
- model
- training
- power equipment
- mask
- rcnn
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10048—Infrared image
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
Translated fromChinese
Description
Translated fromChinese技术领域technical field
本发明涉及一种基于Mask-RCNN的电力设备红外图像分割方法，属于变电站电力设备的图像处理领域。The invention relates to an infrared image segmentation method for power equipment based on Mask-RCNN, which belongs to the field of image processing of power equipment in substations.
背景技术Background technique
近年来，许多监测技术得到积极地推广和应用，在众多监测技术中红外热成像技术因其不停电、不取样、不解体等优点而备受青睐。目前对红外图像的分析诊断多依赖人工，而目前的红外监测人员较少，无法满足对巨量红外图像的分析工作，再加上部分监测人员的专业知识水平和监测经验积累不足，因此对缺陷的分析判断能力较差，这大大制约了设备状态监测智能化水平的提升。In recent years, many monitoring technologies have been actively promoted and applied. Among many monitoring technologies, infrared thermal imaging technology is favored because of its advantages such as no power failure, no sampling, and no disintegration. At present, the analysis and diagnosis of infrared images mostly rely on manual work, and the current infrared monitoring personnel are few, which cannot meet the analysis of huge amounts of infrared images. The ability of analysis and judgment is poor, which greatly restricts the improvement of the intelligent level of equipment condition monitoring.
为了提高电气设备状态监测的智能化水平，可以采用一些智能的图像处理方法对红外图像进行分割，提取出电力设备区域，便于故障诊断工作。在数字图像研究领域，已经出现了很多图像分割方法。目前传统的图像分割方法主要适用于图像目标突出、且非密集的情况，而且分割结果的精确度不高。此外，目前基于传统方法的电力设备分割结果图中，没有保留原始红外图像的图像信息，输出的结果为黑白图像，由于红外图像中的颜色信息对应着设备的温度信息，这是电力设备红外故障诊断的重要依据，因此基于传统方式的电力设备分割方法不利于红外故障诊断分析工作。In order to improve the intelligence level of electrical equipment condition monitoring, some intelligent image processing methods can be used to segment the infrared image and extract the electrical equipment area, which is convenient for fault diagnosis. In the field of digital image research, many image segmentation methods have emerged. At present, the traditional image segmentation methods are mainly suitable for the situation where the image objects are prominent and not dense, and the accuracy of the segmentation results is not high. In addition, the image information of the original infrared image is not retained in the current power equipment segmentation result based on the traditional method, and the output result is a black and white image. Since the color information in the infrared image corresponds to the temperature information of the equipment, this is an infrared fault of the power equipment. Therefore, the traditional power equipment segmentation method is not conducive to the infrared fault diagnosis and analysis work.
发明内容SUMMARY OF THE INVENTION
本发明提供一种基于Mask-RCNN的电力设备红外图像分割方法，克服了所述传统方法的缺点，提高了切割精度，且可以保留设备原始红外图像的图像信息，同直接读取分割后图像中的彩色部分的色彩信息可得到设备温度信息，方便电力设备故障诊断。The invention provides an infrared image segmentation method of power equipment based on Mask-RCNN, which overcomes the shortcomings of the traditional method, improves the cutting accuracy, and can retain the image information of the original infrared image of the equipment, which is the same as directly reading the image after the segmentation. The color information of the color part of the device can get the temperature information of the equipment, which is convenient for the fault diagnosis of the power equipment.
为解决上述技术问题，本发明所采用的技术方案如下：In order to solve the above-mentioned technical problems, the technical scheme adopted in the present invention is as follows:
一种基于Mask-RCNN的电力设备红外图像分割方法，包含以下步骤：A Mask-RCNN-based infrared image segmentation method for power equipment, comprising the following steps:
步骤S1：建立电力设备红外图像的数据集，标注好训练集和测试集；Step S1: establish a data set of infrared images of power equipment, and mark the training set and test set;
步骤S2：构建立深度学习模型；Step S2: build a deep learning model;
步骤S3：设置模型初始超参数和迭代次数，训练集中所有样本被训练模型计算一次就叫做一次迭代；Step S3: Set the initial hyperparameters of the model and the number of iterations. All samples in the training set are calculated once by the training model, which is called one iteration;
步骤S4：使用步骤S1中标注好的训练集，输入构建好的模型中进行训练；Step S4: use the training set marked in step S1, and input it into the constructed model for training;
步骤S5：每2000～3000迭代次数，采用步骤S1中标注好的测试集，评估步骤S4中该次训练得到的模型的性能；Step S5: every 2000-3000 iterations, use the test set marked in step S1 to evaluate the performance of the model obtained by this training in step S4;
步骤S6：当迭代次数达到设定值时，停止训练，综合步骤S5中获取的各个训练阶段的模型超参数，对比各项性能，筛选出性能最优的深度学习模型；Step S6: when the number of iterations reaches the set value, stop training, synthesize the model hyperparameters of each training stage obtained in step S5, compare various performances, and screen out the deep learning model with the best performance;
步骤S7：将待测电力设备红外图像输入训练好最优的深度学习模型进行处理，获得分割结果。Step S7: Input the infrared image of the power equipment to be tested into the optimally trained deep learning model for processing, and obtain a segmentation result.
申请人创造性地将深度学习引入到对电气设备红外图像的分析处理中，申请人经研究发现，深度学习对特征提取过程更加方便智能，无需更多的图像增强等预处理手段就能自动学习得到特征。The applicant creatively introduced deep learning into the analysis and processing of infrared images of electrical equipment. The applicant found through research that deep learning is more convenient and intelligent for the feature extraction process, and can be automatically learned without more preprocessing methods such as image enhancement. feature.
本发明将深度学习引入到对电气设备红外图像的分析处理中，对特征提取过程更加方便智能，无需更多的图像增强等预处理手段就能自动学习得到特征。The invention introduces deep learning into the analysis and processing of infrared images of electrical equipment, which is more convenient and intelligent for the feature extraction process, and can automatically learn to obtain features without more preprocessing means such as image enhancement.
本申请基于Mask-RCNN的电力设备红外图像分割方法，采用智能的图像处理方法对红外图像进行分割，提高了分割精度，且可以保留设备原始红外图像的图像信息，同直接读取分割后图像中的彩色部分的色彩信息可得到设备温度信息，方便电力设备故障诊断。The infrared image segmentation method of power equipment based on Mask-RCNN in this application adopts an intelligent image processing method to segment infrared images, which improves the segmentation accuracy, and can retain the image information of the original infrared images of the equipment. The color information of the color part of the device can get the temperature information of the equipment, which is convenient for the fault diagnosis of the power equipment.
为了进一步提高分割的精度，步骤S1：建立电力设备红外图像的数据集：对红外成像设备在变电站现场采集的红外图像进行预处理，预处理包括统一图像尺寸、及标注出电力设备在图像中的位置和形状，并按(8±2):2比例分为训练集和测试集。In order to further improve the accuracy of segmentation, step S1: establishing a data set of infrared images of power equipment: preprocessing the infrared images collected by infrared imaging equipment at the substation site. The preprocessing includes unifying the image size and marking the power equipment in the image. position and shape, and divided into training set and test set in a ratio of (8±2):2.
优选，步骤S2：采用Mask-RCNN网络架构建立深度学习模型；其中，采用ResNet101和FPN作为特征提取网络。Preferably, step S2: using the Mask-RCNN network architecture to establish a deep learning model; wherein, ResNet101 and FPN are used as the feature extraction network.
为了进一步提高分割的精度，在步骤S3中，迭代次数设置在5000～10000次。同时，在步骤S3中，使用小批量梯度下降算法(MBGD)作为参数更新方式，设置批量大小为32，训练过程中不停地寻找节点中下降幅度最大的趋势进行迭代计算，每次迭代更新一次模型的权重值。In order to further improve the accuracy of segmentation, in step S3, the number of iterations is set at 5000-10000 times. At the same time, in step S3, the mini-batch gradient descent algorithm (MBGD) is used as the parameter update method, and the batch size is set to 32. During the training process, the trend with the largest decline in the nodes is continuously searched for iterative calculation, and each iteration is updated once. The weight value of the model.
优选地，在步骤S3中，若由于条件限制，准备的电力设备红外图像的数据集中数量较少，则使用迁移学习的思想，先使用公开数据集COCO在搭建好的模型中进行预训练，使模型学习到图像分割能力，然后再使用电力设备红外图像的数据集在模型中正式训练；相应地，为防止由于数据集太小引起的过拟合，训练的迭代次数设置在30～50次左右，使用批量梯度下降算法(BGD)作为超参数更新方式，批量大小设置为训练集大小。Preferably, in step S3, if the number of data sets prepared for infrared images of power equipment is small due to conditional constraints, the idea of transfer learning is used, and the public data set COCO is used to pre-train the built model first, so that the The model learns the image segmentation ability, and then uses the data set of infrared images of power equipment to formally train in the model; accordingly, in order to prevent overfitting caused by the data set being too small, the number of training iterations is set at about 30 to 50 times. , using batch gradient descent (BGD) as the hyperparameter update method, and the batch size is set to the training set size.
为了进一步提高分割的精度，步骤S5：每2000～3000迭代次数，采用步骤S1中标注好的测试集，评估步骤S4中该次训练得到的模型的性能；若准确率达到90％(≥90％)，则说明该模型具备较好的分割电力设备红外图像的能力，保存该次训练的超参数文件；若准确率低于90％，则重复步骤S4进行模型的迭代优化。In order to further improve the accuracy of segmentation, step S5: every 2000-3000 iterations, the test set marked in step S1 is used to evaluate the performance of the model obtained by this training in step S4; if the accuracy rate reaches 90% (≥90% ), it means that the model has a good ability to segment infrared images of power equipment, and save the hyperparameter file of this training; if the accuracy rate is lower than 90%, repeat step S4 to iteratively optimize the model.
为了进一步提高精度，步骤S5，评估步骤S4中该次训练得到的模型的性能：准确率、精确率、召回率和平均正确率均达到90％，则说明该模型具备较好的分割电力设备红外图像的能力，保存该次训练的超参数文件；若准确率、精确率、召回率和平均正确率中有一项低于90％，则重复步骤S4进行模型的迭代优化。In order to further improve the accuracy, in step S5, the performance of the model obtained by this training in step S4 is evaluated: the accuracy rate, precision rate, recall rate and average correct rate all reach 90%, indicating that the model has better segmentation power equipment infrared The ability of the image, save the hyperparameter file of the training; if one of the accuracy, precision, recall and average correct rate is lower than 90%, repeat step S4 to iteratively optimize the model.
上述模型的性能包括准确率、精确率、召回率和平均正确率等等评价指标。The performance of the above model includes evaluation indicators such as accuracy, precision, recall, and average accuracy.
本发明未提及的技术均参照现有技术。The technologies not mentioned in the present invention all refer to the prior art.
本发明基于Mask-RCNN的电力设备红外图像分割方法，完成了对红外图像中的电气设备的目标检测和分割，与传统的红外图像分割方法相比，该方法在精度上有很大的提升，并且更加注重目标设备的切割，最大的优势在于该方法保留了目标设备的原色彩信息，将无关的背景和干扰信息完全与目标设备分割开来，可以在未来的红外故障诊断工作中直接读取分割后图像中的彩色部分的色彩信息从而得到温度信息，若温度信息异常，则根据温度信息进一步开展故障诊断工作。The invention based on the Mask-RCNN infrared image segmentation method of electric equipment completes the target detection and segmentation of the electrical equipment in the infrared image. Compared with the traditional infrared image segmentation method, the method has a great improvement in accuracy. And pay more attention to the cutting of the target device. The biggest advantage is that this method retains the original color information of the target device, and completely separates irrelevant background and interference information from the target device, which can be directly read in future infrared fault diagnosis work. The color information of the color part in the segmented image is obtained to obtain temperature information. If the temperature information is abnormal, further fault diagnosis will be carried out according to the temperature information.
附图说明Description of drawings
图1为本发明实施例中基于Mask-RCNN的电力设备红外图像分割方法的流程示意图；1 is a schematic flowchart of an infrared image segmentation method for power equipment based on Mask-RCNN in an embodiment of the present invention;
图2是本发明实施例中Mask R-CNN图像分割算法网络结构示意图；2 is a schematic diagram of the network structure of the Mask R-CNN image segmentation algorithm in the embodiment of the present invention;
图3是本发明实施例中电力设备红外图像样例示意图；3 is a schematic diagram of a sample infrared image of a power device in an embodiment of the present invention;
图4是本发明实施例中电力设备红外待测图像；Fig. 4 is the infrared image to be measured of the electric power equipment in the embodiment of the present invention;
图5是图4的分割结果图。FIG. 5 is a segmentation result diagram of FIG. 4 .
具体实施方式Detailed ways
为了更好地理解本发明，下面结合实施例进一步阐明本发明的内容，但本发明的内容不仅仅局限于下面的实施例。In order to better understand the present invention, the content of the present invention is further illustrated below in conjunction with the embodiments, but the content of the present invention is not limited to the following embodiments.
如图1所示，本实施例方法的整体流程包括以下步骤；As shown in FIG. 1 , the overall flow of the method in this embodiment includes the following steps;
步骤S1；建立电力设备红外图像数据集，图像样本来自于变电站现场的自动巡检机器人、手持红外热成像仪等红外成像设备，选用设备分明、背景清晰和角度正确的红外图像作为训练数据集，图像选取样例如图3所示；红外图像中的电力设备分为三类：电流互感器、电压互感器和断路器，把各类图像的占比调整至约为1:1:1，对筛选出的红外图像进行预处理，首先把尺寸参差不齐的图像统一到256×256像素大小，接着使用图像标注工具VIA标注出电力设备在图像中的位置、形状和类型，完成标注后每张图像会生成同名json格式文件；最后，按8:2比例将预处理后的图像数据集分为训练集和测试集；Step S1; establish an infrared image data set of power equipment, the image samples are from infrared imaging equipment such as automatic inspection robots and handheld infrared thermal imagers on the substation site, and infrared images with clear equipment, clear background and correct angle are selected as the training data set, The image selection sample is shown in Figure 3; the power equipment in the infrared image is divided into three categories: current transformers, voltage transformers and circuit breakers. The infrared images obtained are preprocessed. First, the images with uneven sizes are unified to 256 × 256 pixels, and then the image annotation tool VIA is used to mark the position, shape and type of the power equipment in the image. After completing the annotation, each image is marked. A json format file with the same name will be generated; finally, the preprocessed image dataset is divided into a training set and a test set in an 8:2 ratio;
步骤S2：采用Mask-RCNN网络架构建立深度学习模型；Mask R-CNN作为一个更具有集成能力、更综合和更强大的多任务模型，能够胜任目标检测、图像分类和图像分割的任务；Mask R-CNN有灵活的扩展性，可根据任务的需求选择合适的特征提取网络，根据ResNet残差神经网络和FPN在特征提取过程中有的优秀表现，本例采用ResNet101结合FPN作为特征提取网络，构建的网络结构如图2所示；Step S2: Use the Mask-RCNN network architecture to establish a deep learning model; Mask R-CNN, as a more integrated, more comprehensive and more powerful multi-task model, is capable of target detection, image classification and image segmentation tasks; Mask R -CNN has flexible scalability, and can select a suitable feature extraction network according to the needs of the task. According to the excellent performance of ResNet residual neural network and FPN in the feature extraction process, this example uses ResNet101 combined with FPN as the feature extraction network to construct The network structure is shown in Figure 2;
步骤S3：设置模型初始超参数和迭代次数，设置迭代次数为5000～10000次，参数更新方式为小批量梯度下降算法(MBGD)，批量大小(batch_size)为32；每次迭代，模型对该批量中的图像进行一次计算，通过损失函数的loss值向前反馈和计算超参数差值，对模型的超参数进行更新；Step S3: Set the initial hyperparameters of the model and the number of iterations, set the number of iterations to 5,000 to 10,000 times, the parameter update method is the mini-batch gradient descent algorithm (MBGD), and the batch size (batch_size) is 32; The image in the image is calculated once, and the hyperparameters of the model are updated through the forward feedback of the loss value of the loss function and the calculation of the hyperparameter difference;
步骤S4：使用步骤S1中已进行标注的红外图像训练集，输入构建好的深度模型中进行训练；Step S4: use the infrared image training set that has been marked in step S1, and input it into the constructed deep model for training;
步骤S5：分别在迭代次数达到5000次、8000次、10000次时，采用步骤S1中标注好的测试集，评估步骤S4中该次训练得到的模型的性能，性能具体包括准确率、精确率、召回率、平均正确率等评价指标；Step S5: When the number of iterations reaches 5000 times, 8000 times, and 10000 times respectively, use the test set marked in step S1 to evaluate the performance of the model obtained by this training in step S4, and the performance specifically includes accuracy rate, precision rate, Evaluation indicators such as recall rate and average accuracy rate;
准确率、精确率、召回率和平均正确率均达到90％，则说明当前训练阶段的模型超参数较为适用于当前任务，模型具备较好的分割电力设备红外图像的能力，保存该次训练的超参数文件；The accuracy rate, precision rate, recall rate and average correct rate all reach 90%, which means that the model hyperparameters in the current training stage are more suitable for the current task, and the model has a good ability to segment infrared images of power equipment. hyperparameter file;
若准确率、精确率、召回率和平均正确率中有一项低于90％，则重复步骤S4进行模型的迭代优化；If one of the accuracy rate, precision rate, recall rate and average correct rate is lower than 90%, repeat step S4 to perform iterative optimization of the model;
步骤S6：当迭代次数达到10000次时，停止训练，对比分别在迭代次数为5000次、8000次、10000次时获取的模型性能，筛选出性能最优的一次训练结果，使用该次训练获取的超参数文件，导入模型中作为最终的超参数，得到最优的电力设备红外图像模型。Step S6: When the number of iterations reaches 10,000, stop training, compare the model performance obtained when the number of iterations is 5,000, 8,000, and 10,000, and screen out the training results with the best performance, and use the training results obtained by this training. The hyperparameter file is imported into the model as the final hyperparameter to obtain the optimal infrared image model of power equipment.
步骤S7：选取符合训练集选取条件的电力设备红外图像，作为待测图像输入步骤S6得到的最终的电力设备红外图像模型进行处理，获得分割结果如图4所示,分割结果保留了原图的温度信息，第一行图像为待测图像，第二行图像为分割结果，分割结果包含了电力设备的温度、位置坐标、形状(轮廓)、设备类别(图中，PT为电压互感器)信息。Step S7: Select the infrared image of the power equipment that meets the selection conditions of the training set, and input the final infrared image model of the power equipment obtained in step S6 as the image to be tested. Temperature information, the first line of images is the image to be measured, and the second line of images is the segmentation result. The segmentation result includes the temperature, position coordinates, shape (outline), and equipment category (in the figure, PT is the voltage transformer) information of the power equipment .
在步骤S3中，若由于采集不方便或标注工作繁琐等因素，获取的电力设备红外图像数据集数量较少，则使用迁移学习的思想，先使用公开数据集COCO在搭建好的模型中进行预训练；COCO数据集包含目前自然界的大量其他图像(约12万张)，通过预训练，将模型的超参数先调整到合适的范围，使模型学习到一定的图像分割能力，然后再使用电力设备红外图像数据集在模型中正式训练，这样不仅能解决数据集过小引起的超参数不收敛问题，还能节省训练成本；相应地，为防止由于数据集太小引起的过拟合，减少训练的迭代次数，设置在30～50次左右，使用批量梯度下降算法(BGD)作为超参数更新方式。In step S3, if the number of infrared image data sets of power equipment acquired is small due to factors such as inconvenient collection or cumbersome labeling work, the idea of transfer learning is used, and the public data set COCO is used to pre-predict the built model. Training: The COCO dataset contains a large number of other images (about 120,000 images) in the current nature. Through pre-training, the hyperparameters of the model are adjusted to an appropriate range, so that the model can learn a certain image segmentation ability, and then use electrical equipment. The infrared image data set is formally trained in the model, which can not only solve the problem of non-convergence of hyperparameters caused by the data set being too small, but also save the training cost; accordingly, in order to prevent overfitting caused by the data set being too small, reduce training The number of iterations is set at about 30 to 50 times, and the batch gradient descent algorithm (BGD) is used as the hyperparameter update method.
本例完成了对红外图像中的电气设备的目标检测和分割，与传统的红外图像分割方法相比，该方法在在精度上有很大的提升并且更加注重目标设备的切割，最大的优势在于该方法保留了目标设备的原色彩信息，将无关的背景和干扰信息完全与目标设备分割开来，可以在未来的红外故障诊断工作中直接读取分割后图像中的彩色部分的色彩信息从而得到温度信息，若温度信息异常，则根据温度信息进一步开展故障诊断工作。This example completes the target detection and segmentation of the electrical equipment in the infrared image. Compared with the traditional infrared image segmentation method, this method has a great improvement in accuracy and pays more attention to the cutting of the target equipment. The biggest advantage is that The method retains the original color information of the target device, completely separates irrelevant background and interference information from the target device, and can directly read the color information of the color part of the divided image in the future infrared fault diagnosis work to obtain Temperature information, if the temperature information is abnormal, further fault diagnosis will be carried out according to the temperature information.
Claims (8)
Translated fromChinesePriority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911027493.7ACN110782461A (en) | 2019-10-28 | 2019-10-28 | Mask-RCNN-based infrared image segmentation method for electric power equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911027493.7ACN110782461A (en) | 2019-10-28 | 2019-10-28 | Mask-RCNN-based infrared image segmentation method for electric power equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110782461Atrue CN110782461A (en) | 2020-02-11 |
Family
ID=69386915
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911027493.7APendingCN110782461A (en) | 2019-10-28 | 2019-10-28 | Mask-RCNN-based infrared image segmentation method for electric power equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110782461A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111429412A (en)* | 2020-03-17 | 2020-07-17 | 北京青燕祥云科技有限公司 | Ultrasound AI auxiliary diagnosis method and system for hydatid hepatica |
| CN111768401A (en)* | 2020-07-08 | 2020-10-13 | 中国农业大学 | A rapid freshness classification method of chilled pomfret based on deep learning |
| CN111815568A (en)* | 2020-06-15 | 2020-10-23 | 国网上海市电力公司 | Automatic labeling method and device for power equipment |
| CN112036410A (en)* | 2020-08-13 | 2020-12-04 | 华乘电气科技股份有限公司 | Electric power equipment infrared image segmentation method based on deep learning and image gradient |
| CN112116573A (en)* | 2020-09-16 | 2020-12-22 | 四川嘉能佳网创新能源科技有限责任公司 | A high-precision infrared image anomaly detection method and system |
| CN112733592A (en)* | 2020-11-26 | 2021-04-30 | 中国南方电网有限责任公司超高压输电公司检修试验中心 | Image identification method for power transmission line |
| CN113159334A (en)* | 2021-02-24 | 2021-07-23 | 广西大学 | Electrical equipment infrared image real-time detection and diagnosis method based on light-weight deep learning |
| CN113344475A (en)* | 2021-08-05 | 2021-09-03 | 国网江西省电力有限公司电力科学研究院 | Transformer bushing defect identification method and system based on sequence modal decomposition |
| WO2021248269A1 (en)* | 2020-06-08 | 2021-12-16 | 上海交通大学 | High-precision identification method and system for transformer device |
| CN114034390A (en)* | 2021-11-08 | 2022-02-11 | 山东大学 | A device temperature anomaly detection system based on infrared detection |
| CN114463299A (en)* | 2022-01-26 | 2022-05-10 | 浙江天铂云科光电股份有限公司 | Infrared image detection method for wall bushing |
| CN114943228A (en)* | 2022-06-06 | 2022-08-26 | 北京百度网讯科技有限公司 | Training method of end-to-end sensitive text recall model and sensitive text recall method |
| CN114972193A (en)* | 2022-04-24 | 2022-08-30 | 国电南瑞科技股份有限公司 | Marketing site operation state detection method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CA2995850A1 (en)* | 2015-08-31 | 2017-03-09 | Ryan Kottenstette | Systems and methods for analyzing remote sensing imagery |
| CN107564025A (en)* | 2017-08-09 | 2018-01-09 | 浙江大学 | A kind of power equipment infrared image semantic segmentation method based on deep neural network |
| DE102018120635A1 (en)* | 2017-08-25 | 2019-02-28 | Ford Global Technologies, Llc | DIVIDED PROCESSING WITH DEEP NEURONAL NETWORKS |
- 2019
- 2019-10-28CNCN201911027493.7Apatent/CN110782461A/enactivePending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CA2995850A1 (en)* | 2015-08-31 | 2017-03-09 | Ryan Kottenstette | Systems and methods for analyzing remote sensing imagery |
| CN107564025A (en)* | 2017-08-09 | 2018-01-09 | 浙江大学 | A kind of power equipment infrared image semantic segmentation method based on deep neural network |
| DE102018120635A1 (en)* | 2017-08-25 | 2019-02-28 | Ford Global Technologies, Llc | DIVIDED PROCESSING WITH DEEP NEURONAL NETWORKS |
Non-Patent Citations (2)
| Title |
|---|
| 贺超凯 等: "edX平台教育大数据的学习行为分析与预测", 《中国远程教育》* |
| 鲍新愿: "低对比度情况下太阳能光伏板红外图像的分割算法研究", 《中国优秀博硕士学位论文全文数据库(硕士)工程科技Ⅱ辑》* |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111429412B (en)* | 2020-03-17 | 2023-11-03 | 北京青燕祥云科技有限公司 | Ultrasonic AI auxiliary diagnosis method and system for liver bag worm |
| CN111429412A (en)* | 2020-03-17 | 2020-07-17 | 北京青燕祥云科技有限公司 | Ultrasound AI auxiliary diagnosis method and system for hydatid hepatica |
| US11881015B2 (en) | 2020-06-08 | 2024-01-23 | Shanghai Jiaotong University | High-precision identification method and system for substations |
| WO2021248269A1 (en)* | 2020-06-08 | 2021-12-16 | 上海交通大学 | High-precision identification method and system for transformer device |
| CN111815568A (en)* | 2020-06-15 | 2020-10-23 | 国网上海市电力公司 | Automatic labeling method and device for power equipment |
| CN111768401A (en)* | 2020-07-08 | 2020-10-13 | 中国农业大学 | A rapid freshness classification method of chilled pomfret based on deep learning |
| CN111768401B (en)* | 2020-07-08 | 2024-01-26 | 中国农业大学 | Deep learning-based quick grading method for freshness of ice fresh pomfret |
| CN112036410B (en)* | 2020-08-13 | 2022-06-07 | 华乘电气科技股份有限公司 | Electric power equipment infrared image segmentation method based on deep learning and image gradient |
| CN112036410A (en)* | 2020-08-13 | 2020-12-04 | 华乘电气科技股份有限公司 | Electric power equipment infrared image segmentation method based on deep learning and image gradient |
| CN112116573A (en)* | 2020-09-16 | 2020-12-22 | 四川嘉能佳网创新能源科技有限责任公司 | A high-precision infrared image anomaly detection method and system |
| CN112733592A (en)* | 2020-11-26 | 2021-04-30 | 中国南方电网有限责任公司超高压输电公司检修试验中心 | Image identification method for power transmission line |
| CN113159334B (en)* | 2021-02-24 | 2022-10-11 | 广西大学 | Real-time detection and diagnosis of infrared images of electrical equipment based on lightweight deep learning |
| CN113159334A (en)* | 2021-02-24 | 2021-07-23 | 广西大学 | Electrical equipment infrared image real-time detection and diagnosis method based on light-weight deep learning |
| CN113344475B (en)* | 2021-08-05 | 2021-12-31 | 国网江西省电力有限公司电力科学研究院 | Transformer bushing defect identification method and system based on sequence modal decomposition |
| CN113344475A (en)* | 2021-08-05 | 2021-09-03 | 国网江西省电力有限公司电力科学研究院 | Transformer bushing defect identification method and system based on sequence modal decomposition |
| CN114034390A (en)* | 2021-11-08 | 2022-02-11 | 山东大学 | A device temperature anomaly detection system based on infrared detection |
| CN114034390B (en)* | 2021-11-08 | 2023-11-03 | 山东大学 | Equipment temperature anomaly detection system based on infrared detection |
| CN114463299A (en)* | 2022-01-26 | 2022-05-10 | 浙江天铂云科光电股份有限公司 | Infrared image detection method for wall bushing |
| CN114972193A (en)* | 2022-04-24 | 2022-08-30 | 国电南瑞科技股份有限公司 | Marketing site operation state detection method and system |
| CN114943228A (en)* | 2022-06-06 | 2022-08-26 | 北京百度网讯科技有限公司 | Training method of end-to-end sensitive text recall model and sensitive text recall method |
| CN114943228B (en)* | 2022-06-06 | 2023-11-24 | 北京百度网讯科技有限公司 | End-to-end sensitive text recall model training method, sensitive text recall method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110782461A (en) | Mask-RCNN-based infrared image segmentation method for electric power equipment | |
| CN110598736B (en) | Power equipment infrared image fault positioning, identifying and predicting method | |
| CN113436184B (en) | Power equipment image defect identification method and system based on improved twin network | |
| CN110059694A (en) | The intelligent identification Method of lteral data under power industry complex scene | |
| CN114973002A (en) | Improved YOLOv 5-based ear detection method | |
| CN106991666B (en) | A kind of disease geo-radar image recognition methods suitable for more size pictorial informations | |
| CN109544501A (en) | A kind of transmission facility defect inspection method based on unmanned plane multi-source image characteristic matching | |
| CN114494185B (en) | Electrical equipment fault detection method based on RGB-T multi-scale feature fusion | |
| CN112013966A (en) | Power equipment infrared image processing method based on measured temperature | |
| CN114199381B (en) | Electrical equipment fault detection method for improving infrared detection model | |
| CN113205039B (en) | Power equipment fault image recognition disaster investigation system and method based on multiple DCNN networks | |
| CN107463931A (en) | A kind of real-time pointer instrument reading method and device based on ARM platforms | |
| CN108734691A (en) | A kind of transmission line of electricity defect image recognition methods | |
| CN113326873A (en) | Method for automatically classifying opening and closing states of power equipment based on data enhancement | |
| CN107808375A (en) | Merge the rice disease image detecting method of a variety of context deep learning models | |
| CN118675034A (en) | Bayes-based deep learning remote sensing water quality inversion black and odorous water body identification method | |
| CN112419243B (en) | Power distribution room equipment fault identification method based on infrared image analysis | |
| CN116863237A (en) | Image classification method and device based on category related feature reconstruction | |
| CN115147591A (en) | Transformer equipment infrared image voltage heating type defect diagnosis method and system | |
| CN116091818A (en) | Pointer type instrument reading identification method based on multi-neural network cascading model | |
| CN114399754A (en) | A Character Detection Method of Digital Instrument Based on SW-FSSD | |
| CN112634216B (en) | Insulator self-explosion detection method based on deep learning model | |
| CN118552894A (en) | Substation indoor state identification method and device | |
| CN118503871A (en) | Low-voltage intelligent switchgear fault diagnosis method and system based on deep learning | |
| CN115761489B (en) | A distribution line fire detection method based on YoloV5 network model |
Legal Events
| 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 | ||
| RJ01 | Rejection of invention patent application after publication | Application publication date:20200211 | |
| RJ01 | Rejection of invention patent application after publication |