CN106203453A - A kind of based on compressed sensing biological with abiotic target identification method and system thereof - Google Patents
A kind of based on compressed sensing biological with abiotic target identification method and system thereofDownload PDFInfo
- Publication number
- CN106203453A CN106203453ACN201610571691.XACN201610571691ACN106203453ACN 106203453 ACN106203453 ACN 106203453ACN 201610571691 ACN201610571691 ACN 201610571691ACN 106203453 ACN106203453 ACN 106203453A
- Authority
- CN
- China
- Prior art keywords
- signal
- image
- target
- feature
- biological
- 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.)
- Granted
Links
Classifications
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/10—Image acquisition
- G06V10/12—Details of acquisition arrangements; Constructional details thereof
- G06V10/14—Optical characteristics of the device performing the acquisition or on the illumination arrangements
- G06V10/147—Details of sensors, e.g. sensor lenses
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Vascular Medicine (AREA)
- Image Analysis (AREA)
Abstract
Translated fromChineseDescription
Translated fromChinese技术领域technical field
本发明涉及计算机图像处理技术领域，特别是涉及一种基于压缩感知的生物与非生物目标识别方法及其系统。The invention relates to the technical field of computer image processing, in particular to a method and system for identifying biological and non-biological targets based on compressed sensing.
背景技术Background technique
目前，海洋摄影数据的获取是利用成像技术进行长周期、大范围、高精度成像，通过图像存储和传输系统，结合数据处理和分析方法，对生物指标及其它关注目标进行分析和分类，其数据存储、传输成本较大，而且处理速度受限，直接影响对海洋生态的评估与预测，所以传统先成像再分类方法存在着速度慢成本高的缺点。现有的压缩感知成像系统主要目的是完整复原图像，因此测量矩阵需要满足RIP条件，应用的随机高斯矩阵、Toplize(托普利茨)矩阵、伯努利矩阵或者各行非相关的确定性矩阵较多，图像复原也依据所选用的矩阵进行复原。At present, the acquisition of marine photography data is to use imaging technology for long-term, large-scale, high-precision imaging, through image storage and transmission systems, combined with data processing and analysis methods, to analyze and classify biological indicators and other targets of concern. The cost of storage and transmission is high, and the processing speed is limited, which directly affects the assessment and prediction of marine ecology. Therefore, the traditional method of imaging first and then classifying has the disadvantage of slow speed and high cost. The main purpose of the existing compressed sensing imaging system is to completely restore the image, so the measurement matrix needs to meet the RIP condition, and the applied random Gaussian matrix, Toplize (Toplitz) matrix, Bernoulli matrix or non-correlated deterministic matrix of each row is relatively More, the image restoration is also restored according to the selected matrix.
综上，海洋摄影数据的成像技术如果能运用压缩感知的进行获取，基于压缩感知的技术将大大减少数据量的存储与硬件的消耗，提高传输速度以及降低成本。To sum up, if the imaging technology of marine photography data can be obtained by using compressed sensing, the technology based on compressed sensing will greatly reduce the amount of data storage and hardware consumption, improve transmission speed and reduce costs.
但是，现有的海洋摄影数据主要是以图像恢复为目的来进行海洋生物以及非生物数据的获取，以完整复原图像为目的的传统压缩感知技术，是在图像复原以后对生物及非生物的数据进行识别，没能在采样的过程中进行识别，而实际上海洋摄影数据最终的目的是识别出生物及非生物特征，所以导致以完整复原图像为目的的传统压缩感知技术也不能很好的减轻海洋摄影数据成像技术进行长周期、大范围、高精度成像的负担，无法直接分类识别出生物与非生物特征，导致无法快速的获取想要的数据结果。However, the existing marine photography data is mainly for the purpose of image restoration to obtain marine biological and non-biological data. The traditional compressed sensing technology for the purpose of complete restoration of images is to recover the biological and non-biological data after image restoration. In fact, the ultimate purpose of marine photography data is to identify biological and non-biological features, so the traditional compressed sensing technology for the purpose of completely restoring the image cannot be well alleviated. The burden of long-period, large-scale, and high-precision imaging for marine photography data imaging technology cannot directly classify and identify biological and non-biological features, resulting in the inability to quickly obtain the desired data results.
并且，如果要运用压缩感知技术快速识别生物特征及非生物特征，由于基于海洋生物及非生物特征所获得测量矩阵的矩阵模式为确定性矩阵，不能符合RIP条件，无法满足各行非相关，所以采用以完整复原图像为目的的传统压缩感知技术进行特征识别便无法进行，尤其是在图像复原的过程中尤为困难。Moreover, if compressed sensing technology is to be used to quickly identify biological and non-biological features, since the matrix mode of the measurement matrix obtained based on marine biological and non-biological features is a deterministic matrix, it cannot meet the RIP conditions and cannot satisfy the non-correlation of each row, so use The traditional compressed sensing technology for the purpose of completely restoring the image cannot be used for feature recognition, especially in the process of image restoration.
以上背景技术内容的公开仅用于辅助理解本发明的发明构思及技术方案，其并不必然属于本专利申请的现有技术，在没有明确的证据表明上述内容在本专利申请的申请日已经公开的情况下，上述背景技术不应当用于评价本申请的新颖性和创造性。The disclosure of the above background technical content is only used to assist in understanding the inventive concepts and technical solutions of the present invention, and it does not necessarily belong to the prior art of this patent application. There is no clear evidence that the above content has been disclosed on the filing date of this patent application. Under the circumstances, the above background technology should not be used to evaluate the novelty and inventiveness of this application.
发明内容Contents of the invention
本发明目的在于提出一种基于压缩感知的生物与非生物目标识别方法及其系统，以解决上述现有技术存在的数据传输速度慢、成本高，其压缩负担重、不适用于目标识别的技术问题。The purpose of the present invention is to propose a biological and non-biological target recognition method and system based on compressed sensing to solve the problems of slow data transmission, high cost, heavy compression burden and unsuitable technology for target recognition in the above-mentioned prior art. question.
为此，本发明提出一种基于压缩感知的生物与非生物目标识别方法，包括以下步骤：S1：通过成像光路模块对目标进行成像，将目标图像投影于孔径编码器上；S2：信号控制模块根据预定的特征识别模式生成相对应的确定矩阵模式，通过所述确定矩阵模式控制所述孔径编码器得到至少一个待处理特征信号，完成目标图像的压缩计算及特征识别；S3：所述待处理特征信号经图像处理及分类模块负责完成放大、处理和训练比较分类，得到最终特征信号。For this reason, the present invention proposes a biological and non-biological target recognition method based on compressed sensing, including the following steps: S1: Imaging the target through the imaging optical path module, and projecting the target image on the aperture encoder; S2: Signal control module Generate a corresponding definite matrix mode according to a predetermined feature recognition mode, control the aperture encoder through the definite matrix mode to obtain at least one feature signal to be processed, and complete the compression calculation and feature recognition of the target image; S3: the pending process The image processing and classification module of the characteristic signal is responsible for completing amplification, processing, training, comparison and classification to obtain the final characteristic signal.
优选地，本发明还可以具有如下技术特征：Preferably, the present invention can also have the following technical features:
步骤S1中，包括根据预定的特征识别模式对目标图像进行灰度处理，将经灰度处理获得的灰度图像投影于所述孔径编码器上。Step S1 includes performing grayscale processing on the target image according to a predetermined feature recognition pattern, and projecting the grayscale image obtained through the grayscale processing on the aperture encoder.
步骤S2中，该确定矩阵模式对应于所述特征识别模式下特征识别运算的算子，所述灰度图像经所述算子卷积，生成至少一个待处理特征信号。In step S2, the determined matrix mode corresponds to an operator of a feature recognition operation in the feature recognition mode, and the grayscale image is convoluted by the operator to generate at least one feature signal to be processed.
步骤S3中，所述图像处理及分类模块包括经训练得到的生物特征库，所述待处理特征信号通过所述生物特征库对生物特征进行比较分类，并通过直方图将所述待处理特征信号进行识别。In step S3, the image processing and classification module includes a trained biological feature library, and the feature signal to be processed compares and classifies the biological features through the biological feature library, and compares and classifies the feature signal to be processed through a histogram to identify.
步骤S2中，所述特征识别模式包括基于HOG特征、LBP特征、Haar特征或Curvelet变换的特征识别。In step S2, the feature recognition mode includes feature recognition based on HOG features, LBP features, Haar features or Curvelet transform.
包括S11：对所述灰度图像进行分区，得到包括多个像素的细胞单元，并由多个所述细胞单元构成一区块；S12：根据所述区块的像素密度，对所述区块内的各细胞单元作归一化处理，以得到所述目标图像。Including S11: partitioning the grayscale image to obtain a cell unit including a plurality of pixels, and forming a block by a plurality of the cell units; S12: according to the pixel density of the block, dividing the block Each cell unit within is subjected to normalization processing to obtain the target image.
包括S21：所述信号控制模块根据水平梯度算子Gx控制所述孔径编码器对目标图像的多个像素点作为采样点进行多次采样，得到水平梯度信号X；S22、所述信号控制模块根据垂直梯度算子Gy控制所述孔径编码器对目标图像的多个像素点作为采样点进行多次采样，得到垂直梯度信号Y。Including S21: the signal control module controls the aperture encoder to perform multiple samplings on multiple pixels of the target image as sampling points according to the horizontal gradient operatorGx to obtain a horizontal gradient signal X; S22, the signal control module The aperture encoder is controlled according to the vertical gradient operator Gy to perform multiple samplings on multiple pixels of the target image as sampling points to obtain a vertical gradient signal Y.
步骤S3中，所述图像处理及分类模块的处理器根据水平梯度信号X和所述垂直梯度信号Y进行放大，根据所述细胞单元生成第一方向梯度直方图，根据所述第一方向梯度直方图生成所述区块的第二方向梯度直方图，根据所述第二方向梯度直方图生成目标图像的第三方向梯度直方图，所述根据所述细胞单元生成第一方向梯度直方图，包括根据训练后的生物特征库对所述第一方向梯度直方图进行权重投影，完成特征的训练分类。In step S3, the processor of the image processing and classification module amplifies according to the horizontal gradient signal X and the vertical gradient signal Y, generates a gradient histogram in the first direction according to the cell unit, and generates a gradient histogram in the first direction according to the gradient histogram in the first direction. Generate the second directional gradient histogram of the block, generate the third directional gradient histogram of the target image according to the second directional gradient histogram, and generate the first directional gradient histogram according to the cell unit, including Weight projection is performed on the gradient histogram in the first direction according to the trained biological feature database to complete feature training and classification.
本发明还提供了一种基于压缩感知的生物与非生物特征的目标识别系统，用于实现上述的识别方法，包括成像光路模块、信号控制模块、图像处理及分类模块，所述成像光路模块包括第一透镜、孔径编码器、反射镜、第二透镜、光电传感器，所述信号控制模块包括信号发生器，所述图像处理及分类模块包括处理器、放大器和生物特征库。The present invention also provides a biological and non-biological target recognition system based on compressed sensing, which is used to realize the above-mentioned recognition method, including an imaging optical path module, a signal control module, an image processing and classification module, and the imaging optical path module includes The first lens, an aperture encoder, a mirror, a second lens, and a photoelectric sensor, the signal control module includes a signal generator, and the image processing and classification module includes a processor, an amplifier, and a biometric library.
所述孔径编码器为数字微镜器件、随机反射镜、可变形镜或复合孔径掩膜。The aperture encoder is a digital micromirror device, a random mirror, a deformable mirror or a composite aperture mask.
本发明与现有技术对比的有益效果包括：本发明的基于特征的压缩感知目标识别方法，主要目的是为了目标识别，相比于传统的压缩感知技术而言，能在采样的过程中实现特征的识别，以直接识别特征，不再需要完全恢复原图，适用于海洋摄影数据的提取，本发明其测量矩阵的模式为确定性矩阵，为了能通过该矩阵模式进行特征识别，本发明采用了按预定的特征识别模式生成相对应的确定矩阵以作为该压缩感知的测量矩阵，通过孔径编码器根据生成的对应的测量矩阵将目标图像中的特征识别出来，同时完成目标图像的数据压缩，这样本发明能够通过信号控制模块和孔径编码器便在采样的过程中对图像进行压缩的同时，直接获得识别的特征信号，所以相比于现有的海洋摄影数据技术相比，本发明将目标识别以及压缩作为主要目的，不用完成对图像的恢复，使得海洋摄影的负担进一步降低，大大的降低了海洋摄影的成本，减轻了硬件的消耗，提高了图像的传输速度。本发明的测量矩阵(确定矩阵)运用预定的特征识别模式下生成的矩阵模式用于特征的提取，通过孔径编码器实现的在采样的过程的压缩与特征识别，可直接得到该特征结果，所以使得基于确定性矩阵的压缩感知技术能得以很好的运用，为海洋成像观测提供新的手段和工具。The beneficial effects of the present invention compared with the prior art include: the feature-based compressed sensing target recognition method of the present invention, the main purpose is for target recognition, compared with the traditional compressed sensing technology, the feature can be realized in the process of sampling It is suitable for the extraction of marine photographic data. The mode of the measurement matrix of the present invention is a deterministic matrix. In order to perform feature recognition through the matrix mode, the present invention adopts According to the predetermined feature recognition mode, the corresponding determination matrix is generated as the measurement matrix of the compressed sensing, and the features in the target image are identified through the aperture encoder according to the generated corresponding measurement matrix, and the data compression of the target image is completed at the same time, so that The present invention can compress the image during the sampling process through the signal control module and the aperture encoder, and directly obtain the characteristic signal for identification, so compared with the existing marine photographic data technology, the present invention will target recognition As well as compression as the main purpose, there is no need to complete the restoration of the image, which further reduces the burden of marine photography, greatly reduces the cost of marine photography, reduces hardware consumption, and improves image transmission speed. The measurement matrix (determining matrix) of the present invention uses the matrix pattern generated under the predetermined feature recognition mode for feature extraction, and the feature result can be directly obtained through the compression and feature recognition in the sampling process realized by the aperture encoder, so The compressed sensing technology based on the deterministic matrix can be well used, providing new means and tools for ocean imaging observation.
优选方案中，在步骤S1中，包括根据预定的特征识别模式对目标图像进行灰度处理，步骤S2中，该确定矩阵模式对应于所述特征识别模式下特征识别运算的算子，步骤S3中，所述图像处理及分类模块包括经训练得到的生物特征库，并通过直方图将所述待处理特征信号进行识别，灰度处理可以更好地进行图像识别，运用算子为图像识别在孔径编码器上更好的实现，运用生物特征库为图像分类提供更多便捷，通过直方图能直观的将特征表示出来。In a preferred solution, in step S1, it includes performing grayscale processing on the target image according to a predetermined feature recognition mode, in step S2, the determined matrix mode corresponds to the operator of the feature recognition operation in the feature recognition mode, in step S3 , the image processing and classification module includes a biometric library obtained through training, and the characteristic signal to be processed is identified through a histogram, grayscale processing can better perform image recognition, and the operator is used for image recognition in the aperture Better implementation on the encoder, using the biometric library to provide more convenience for image classification, and the features can be intuitively expressed through the histogram.
优选方案中，提供了包括HOG特征、LBP特征、Haar特征或Curvelet变换的特征识别模式，各自结合压缩感知技术，可以对生物特征进行不同模式的表示。In a preferred solution, feature recognition modes including HOG features, LBP features, Haar features or Curvelet transformation are provided, each of which is combined with compressed sensing technology to represent biological features in different modes.
本发明提出了的一种基于压缩感知的生物与非生物特征的识别系统，通过孔径编码器结合信号控制模块实现了图像的压缩及识别，再经图像处理及分类模块完成特征的处理及分类。特别的，系统中为了适应不同的矩阵模式，可以将所述孔径编码器设置为数字微镜器件、随机反射镜、可变形镜或复合孔径掩膜，已完成不同特征识别模式下的特征识别。The present invention proposes a biological and non-biological feature recognition system based on compressed sensing, which realizes image compression and recognition through an aperture encoder combined with a signal control module, and then completes feature processing and classification through an image processing and classification module. In particular, in order to adapt to different matrix modes in the system, the aperture encoder can be set as a digital micromirror device, a random mirror, a deformable mirror or a composite aperture mask to complete feature recognition in different feature recognition modes.
附图说明Description of drawings
图1是本发明具体实施方式一的方法流程图；Fig. 1 is the method flowchart of embodiment one of the present invention;
图2是本发明具体实施方式一的灰度处理的目标图像；Fig. 2 is the target image of the grayscale processing of Embodiment 1 of the present invention;
图3是本发明具体实施方式一区块的HOG特征方向梯度直方图；Fig. 3 is the HOG feature direction gradient histogram of a block in the specific embodiment of the present invention;
图4是本发明具体实施方式一目标图像的HOG特征的方向梯度直方图；Fig. 4 is the directional gradient histogram of the HOG feature of the specific embodiment of the present invention one target image;
图5是本发明具体实施方式一系统的结构组成示意图。Fig. 5 is a schematic diagram of the structural composition of a system according to a specific embodiment of the present invention.
1-目标图像，2-第一透镜，3-反射镜，4-第二透镜，5-光电传感器，6-放大器，7-处理器，8-图像处理及分类模块，9-信号控制模块，10-孔径编码器，11-成像光路模块。1-target image, 2-first lens, 3-mirror, 4-second lens, 5-photoelectric sensor, 6-amplifier, 7-processor, 8-image processing and classification module, 9-signal control module, 10-aperture encoder, 11-imaging optical path module.
具体实施方式detailed description
下面结合具体实施方式并对照附图对本发明作进一步详细说明。应该强调的是，下述说明仅仅是示例性的，而不是为了限制本发明的范围及其应用。The present invention will be further described in detail below in combination with specific embodiments and with reference to the accompanying drawings. It should be emphasized that the following description is only exemplary and not intended to limit the scope of the invention and its application.
参照以下附图1-5，将描述非限制性和非排他性的实施例，其中相同的附图标记表示相同的部件，除非另外特别说明。Non-limiting and non-exclusive embodiments will be described with reference to the following Figures 1-5, wherein like reference numerals refer to like parts unless specifically stated otherwise.
实施例一：Embodiment one:
已知的压缩感知系统，对于一维信号，压缩感知的数学过程描述为：For known compressed sensing systems, for one-dimensional signals, the mathematical process of compressed sensing is described as:
y＝Φxy=Φx
其中in
y是测量的信号，x是待测信号，x1为一维信号n个采样点的第一个采样点，可视为像素，xn为第n个采样点；y1为收集到m个压缩后的一维信号中第一个信号，ym为第m个信号；y is the measured signal, x is the signal to be measured, x1 is the first sampling point of n sampling points of the one-dimensional signal, which can be regarded as a pixel, xn is the nth sampling point; y1 is the collected m The first signal in the compressed one-dimensional signal, ym is the mth signal;
Φ是测量矩阵，Φ11-Φ1n为测量矩阵的第1个测量模式下的n个测量值，相应的Φm1-Φmn为测量矩阵的第m个测量模式下的n个测量值，例如在数字微镜器件(DMD)中，Φ11-Φ1n为第一次的DMD面元阵列表达。Φ is the measurement matrix, Φ11 -Φ1n are the n measured values in the first measurement mode of the measurement matrix, and the corresponding Φm1 -Φmn are the n measured values in the mth measurement mode of the measurement matrix, for example In the digital micromirror device (DMD), Φ11 -Φ1n is the first expression of DMD surface element array.
Φ需要满足RIP(有限等距性)条件，一般需要选择Φ的各行不相关，待测信号可以表示为x＝Ψs，在某个稀疏基Ψ下，信号x可以表达为k个重要分量的组成形式(k<n)，即表示x的稀疏表达s有k个不为0的分量，则y＝ΦΨs，除此之外，对于测量矩阵为随机高斯矩阵的形式，测量次数m需要满足m≈4k或者m≥Klog(n/k)。Φ needs to satisfy the RIP (finite isometric) condition. Generally, the rows of Φ need to be selected to be irrelevant. The signal to be tested can be expressed as x=Ψs. Under a certain sparse basis Ψ, the signal x can be expressed as the composition of k important components The form (k<n), which means that the sparse expression s of x has k components that are not 0, then y=ΦΨs, in addition, for the form of the measurement matrix is a random Gaussian matrix, the number of measurements m needs to satisfy m≈ 4k or m≥Klog(n/k).
在上述条件下，信号几乎可以完全恢复，基于特征识别的数学表达类似于上述表达，如图1所示，本实施例提出的一种基于压缩感知的生物与非生物目标识别方法，包括以下步骤：S1：通过成像光路模块11对目标进行成像，将目标图像1投影于孔径编码器10上；S2：所述信号控制模块9根据预定的特征识别模式生成相对应的确定矩阵模式，通过所述确定矩阵模式控制所述孔径编码器10得到至少一个待处理特征信号，完成目标图像1的压缩计算及特征识别；S3：所述待处理特征信号经图像处理及分类模块8负责完成放大、处理和训练比较分类，得到最终特征信号。Under the above conditions, the signal can be recovered almost completely, and the mathematical expression based on feature recognition is similar to the above expression, as shown in Figure 1, a method for identifying biological and non-biological targets based on compressed sensing proposed in this embodiment includes the following steps : S1: the target is imaged by the imaging optical path module 11, and the target image 1 is projected on the aperture encoder 10; S2: the signal control module 9 generates a corresponding determination matrix mode according to a predetermined feature recognition mode, through the Determine the matrix mode to control the aperture encoder 10 to obtain at least one feature signal to be processed, and complete the compression calculation and feature recognition of the target image 1; S3: the image processing and classification module 8 of the feature signal to be processed is responsible for completing amplification, processing and The training compares the classification and obtains the final feature signal.
同时为实现上述方法，如图5所示，提供了一种基于压缩感知的生物与非生物特征的识别系统，包括成像光路模块11、信号控制模块9、图像处理及分类模块8，所述成像光路模块11包括第一透镜2、孔径编码器10、反射镜3、第二透镜4、光电传感器5，所述信号控制模块9包括信号发生器，所述图像处理及分类模块8包括处理器7、放大器6和生物特征库。所述孔径编码器10为数字微镜器件、随机反射镜、可变形镜或复合孔径掩膜。其中关键部分为基于特征的压缩感知的测量矩阵的硬件实现，本系统中，测量矩阵的实现为孔径编码器10，根据不同的特征识别模式，该孔径编码器10可以为数字微镜器件(Digital Mirror Device,DMD)、随机反射镜、可变形镜或复合孔径掩膜，以数字微镜器件为例，DMD的铰链单元可以实现(-12°，12°)，可以用于测量矩阵为伯努利矩阵的压缩感知系统，而可变形镜由于其面形可以随电压可连续变化，因而可以产生更多的模式，测量矩阵不局限于伯努利矩阵，为设计基于特征的测量矩阵提供了更多的可能性。Simultaneously for realizing above-mentioned method, as shown in Figure 5, provide a kind of identification system based on compressive sensing biological and non-biological features, including imaging optical path module 11, signal control module 9, image processing and classification module 8, described imaging The optical path module 11 includes a first lens 2, an aperture encoder 10, a mirror 3, a second lens 4, and a photoelectric sensor 5, the signal control module 9 includes a signal generator, and the image processing and classification module 8 includes a processor 7 , amplifier 6, and biometric library. The aperture encoder 10 is a digital micromirror device, a random mirror, a deformable mirror or a composite aperture mask. Wherein the key part is the hardware implementation of the measurement matrix based on the compressed sensing of features. In this system, the realization of the measurement matrix is an aperture encoder 10. According to different feature recognition modes, the aperture encoder 10 can be a digital micromirror device (Digital Mirror Device, DMD), random mirror, deformable mirror or composite aperture mask, taking digital micromirror device as an example, the hinge unit of DMD can realize (-12°, 12°), and can be used to measure the matrix as Bernoulli The compressive sensing system of the benefit matrix, and the deformable mirror can generate more modes because its surface shape can change continuously with the voltage. many possibilities.
在目标识别的过程中，步骤S1中，包括根据预定的特征识别模式对目标图像1进行灰度处理，将经灰度处理获得的灰度图像投影于所述孔径编码器10上。In the target recognition process, step S1 includes performing grayscale processing on the target image 1 according to a predetermined feature recognition pattern, and projecting the grayscale image obtained through the grayscale processing on the aperture encoder 10 .
步骤S2中，该确定矩阵模式对应于所述特征识别模式下特征识别运算的算子，所述灰度图像经所述算子卷积，生成至少一个待处理特征信号。In step S2, the determined matrix mode corresponds to an operator of a feature recognition operation in the feature recognition mode, and the grayscale image is convoluted by the operator to generate at least one feature signal to be processed.
步骤S3中，所述图像处理及分类模块8包括经训练得到的生物特征库，所述待处理特征信号通过所述生物特征库对生物特征进行比较分类，并通过直方图将所述待处理特征信号进行识别。In step S3, the image processing and classification module 8 includes a trained biological feature library, and the feature signal to be processed compares and classifies the biological features through the biological feature library, and compares and classifies the biological features through the histogram. The signal is identified.
上述的特征识别模式模式包括基于HOG(Histogram of Oriented Gradient,HOG，方向梯度直方图)特征、LBP(Local Binary Patterns，局部二值模式)特征、Haar(Haar-like features，哈尔特征)特征或Curvelet变换(曲波变换)的特征识别。The above-mentioned feature recognition patterns include HOG (Histogram of Oriented Gradient, HOG, directional gradient histogram) features, LBP (Local Binary Patterns, local binary pattern) features, Haar (Haar-like features, Haar feature) features or Feature recognition of Curvelet transform (curvelet transform).
本实施例中，选择HOG特征作为生物及非生物分类识别的主要特征，并对其作更为具体的说明。In this embodiment, the HOG feature is selected as the main feature for classification and identification of organisms and non-organisms, and a more specific description will be given.
步骤S1中，包括S11：对所述灰度图像进行分区，得到包括多个像素的细胞单元，并由多个所述细胞单元构成一区块；S12：根据所述区块的像素密度，对所述区块内的各细胞单元作归一化处理，以得到所述目标图像1。In step S1, including S11: partitioning the grayscale image to obtain a cell unit including a plurality of pixels, and forming a block from a plurality of the cell units; S12: according to the pixel density of the block, Each cell unit in the block is normalized to obtain the target image 1 .
步骤S2中，包括S21：所述信号控制模块9根据水平梯度算子Gx控制所述孔径编码器10对目标图像1的多个像素点作为采样点进行多次采样，得到水平梯度信号X；S22、所述信号控制模块9根据垂直梯度算子Gy控制所述孔径编码器10对目标图像1的多个像素点作为采样点进行多次采样，得到垂直梯度信号Y。In step S2, including S21: the signal control module 9 controls the aperture encoder 10 to perform multiple samplings on multiple pixels of the target image 1 as sampling points according to the horizontal gradient operatorGx , to obtain a horizontal gradient signal X; S22. The signal control module 9 controls the aperture encoder 10 to perform multiple samplings on multiple pixels of the target image 1 as sampling points according to the vertical gradient operator Gy to obtain a vertical gradient signal Y.
步骤S3中，所述图像处理及分类模块8的处理器7根据水平梯度信号X和所述垂直梯度信号Y进行放大，根据所述细胞单元生成第一方向梯度直方图，根据所述第一方向梯度直方图生成所述区块的第二方向梯度直方图，根据所述第二方向梯度直方图生成目标图像1的第三方向梯度直方图，所述根据所述细胞单元生成第一方向梯度直方图，包括根据训练后的生物特征库对所述第一方向梯度直方图进行权重投影，完成特征的训练分类。In step S3, the processor 7 of the image processing and classification module 8 amplifies according to the horizontal gradient signal X and the vertical gradient signal Y, generates a gradient histogram in the first direction according to the cell unit, and generates a gradient histogram in the first direction according to the first direction The gradient histogram generates the second direction gradient histogram of the block, generates the third direction gradient histogram of the target image 1 according to the second direction gradient histogram, and generates the first direction gradient histogram according to the cell unit The figure includes performing weight projection on the gradient histogram in the first direction according to the trained biological feature database to complete feature training and classification.
如上所述，HOG特征计算过程将图像分为多个区块，每个区块包括几个小的区域(细胞单元)，通过计算每个小区域所有像素的方向梯度直方图，组成各个区块的方向梯度直方图，最后组成整幅图像的方向梯度直方图。其中最关键的运算是求每个小区域中各像素的垂直梯度和水平梯度，对应的算子是Gx＝[-1,0,1],Gy＝[1,0,-1]T，孔径编码器10需要x方向与y方向各重复3次运算，可以完成所有的求梯度运算，然后通过对记录对应位置的电压信号进行放大和AD转换，得到相应数字信号，通过处理器7以及生物特征库进行处理后，完成一幅图像的HOG特征产生过程。As mentioned above, the HOG feature calculation process divides the image into multiple blocks, and each block includes several small areas (cell units). By calculating the direction gradient histogram of all pixels in each small area, each block is composed The directional gradient histogram of , and finally the directional gradient histogram of the entire image. The most critical operation is to calculate the vertical gradient and horizontal gradient of each pixel in each small area, and the corresponding operator is Gx ＝[-1,0,1], Gy ＝[1,0,-1]T , the aperture encoder 10 needs to repeat the operation 3 times in the x direction and the y direction, and can complete all the gradient calculations, and then amplify and AD convert the voltage signal at the corresponding position to obtain the corresponding digital signal, through the processor 7 and After the biometric database is processed, the HOG feature generation process of an image is completed.
若图像可表示为数学上的梯度计算在压缩感知的数学公式表达为：If the image can be expressed as The mathematical gradient calculation is expressed in the mathematical formula of compressed sensing as:
公式中，矩阵I中的行(Im1-Imn)为图像x方向的像素值，列(I1n-Imn)为图像y方向的像素值，共m行、n列，垂直梯度φx为垂直梯度算子Gx与矩阵I的卷积，垂直梯度φy为垂直梯度算子Gy与矩阵I的卷积(conv)。In the formula, the row (Im1 -Imn ) in the matrix I is the pixel value in the x direction of the image, and the column (I1n -Imn ) is the pixel value in the y direction of the image, with m rows and n columns in total, and the vertical gradient φx is Convolution of vertical gradient operator Gx and matrix I, vertical gradient φy is convolution (conv) of vertical gradient operator Gy and matrix I.
如图2所示，为一张像素尺寸为2448x2050的图像，其中的生物特征为水母，本实施例中，将以这张水母图进行举例说明。As shown in FIG. 2 , it is an image with a pixel size of 2448×2050, in which the biological feature is a jellyfish. In this embodiment, this jellyfish image will be used as an example for illustration.
经过灰度处理的目标图像1(水母图)经所述成像光路模块11的第一透镜2投射至孔径编码器10，信号控制模块9的信号发生器根据HOG特征的算子Gx＝[-1,0,1],Gy＝[1,0,-1]T将所述孔径编码器10生成对应的矩阵模式，根据所述矩阵模式多次对压缩图像进行压缩和特征识别，得到的待处理特征信号，经反射镜3、第二透镜4和光电传感器5以及放大器6进行处理，生物特征经过训练过的生物特征库进行分类。The target image 1 (jellyfish figure) processed through the grayscale is projected to the aperture encoder 10 through the first lens 2 of the imaging optical path module 11, and the signal generator of the signal control module 9 is according to the operator Gx of the HOG feature=[- 1,0,1], Gy =[1,0,-1]T generates the corresponding matrix pattern by the aperture encoder 10, performs compression and feature recognition on the compressed image multiple times according to the matrix pattern, and obtains The characteristic signal to be processed is processed through the mirror 3, the second lens 4, the photoelectric sensor 5 and the amplifier 6, and the biological characteristics are classified through the trained biological characteristic library.
运用HOG特征，需要分别对图像的垂直梯度，和水平梯度进行计算，孔径编码器(DMD)将水母图转换成一个二维矩阵此处m为水母图的水平方向x的像素个数2448，n为垂直方向y的像素个数2050。Using the HOG feature, it is necessary to calculate the vertical gradient and horizontal gradient of the image separately, and the aperture encoder (DMD) converts the jellyfish image into a two-dimensional matrix Here m is the number of pixels in the horizontal direction x of the jellyfish image, 2448, and n is the number of pixels in the vertical direction y, 2050.
如图3所示，图中包括6幅图，六张图中，它们各自的x坐标轴和y坐标轴表示图像的尺寸大小，等同于DMD的阵列，z坐标表示DMD的模式选择，其中有两个值，0或1。0或1对应梯度算子的0和1。As shown in Figure 3, there are 6 pictures in the picture. In the six pictures, their respective x-axis and y-coordinate axis represent the size of the image, which is equivalent to the array of DMD. The z coordinate represents the mode selection of DMD, of which there are two Value, 0 or 1. 0 or 1 corresponds to 0 and 1 of the gradient operator.
图3中第一行的3张图表示x方向的梯度算子扫描了3次后得到的方向梯度直方图，第二行的3张图表示y方向的梯度算子扫描了3次后得到的方向梯度直方图，这6次计算完成了以往需要在后端的梯度计算，减少了很大的计算量。The three pictures in the first row in Figure 3 represent the directional gradient histogram obtained after scanning the gradient operator in the x direction three times, and the three pictures in the second row represent the gradient operator in the y direction obtained after scanning three times. Oriented gradient histogram, these 6 calculations completed the gradient calculation that was required in the backend in the past, reducing a lot of calculation.
图4中，同样具有6幅图，图4为在图3的空间编码分布下，水母图像矩阵I与孔径编码对应的矩阵(垂直算子Gy和水平算子Gx)的乘积，同样，x坐标轴和y坐标轴表示图像的尺寸大小，此处z坐标表示处理后水母的像素值。In Fig. 4, there are 6 pictures equally, and Fig. 4 is under the spatial coding distribution of Fig. 3, the product of jellyfish image matrix I and the corresponding matrix (vertical operator Gy and horizontal operator Gx ) of aperture coding, same, The x-coordinate axis and the y-coordinate axis represent the size of the image, and the z coordinate represents the pixel value of the jellyfish after processing.
本实施例中，孔径编码器的硬件计算，可将上部分提到的计算通过物理硬件方式进行计算，采用物理硬件的方式进行计算，极大了减轻了本来需有软件和算法对特征进行提取计算识别的巨大负担，大大提高海洋摄影识别的效率，减轻成本。In this embodiment, the hardware calculation of the aperture encoder can be calculated by means of physical hardware, which greatly reduces the need for software and algorithms to extract features. The huge burden of calculation and identification greatly improves the efficiency of marine photography identification and reduces the cost.
本领域技术人员将认识到，对以上描述做出众多变通是可能的，所以实施例仅是用来描述一个或多个特定实施方式。Those skilled in the art will recognize that many variations on the above description are possible, so the examples are merely intended to describe one or more particular implementations.
尽管已经描述和叙述了被看作本发明的示范实施例，本领域技术人员将会明白，可以对其作出各种改变和替换，而不会脱离本发明的精神。另外，可以做出许多修改以将特定情况适配到本发明的教义，而不会脱离在此描述的本发明中心概念。所以，本发明不受限于在此披露的特定实施例，但本发明可能还包括属于本发明范围的所有实施例及其等同物。Although there have been described and described what are considered to be exemplary embodiments of the present invention, it will be apparent to those skilled in the art that various changes and substitutions may be made therein without departing from the spirit of the invention. In addition, many modifications may be made to adapt a particular situation to the teachings of the invention without departing from the inventive central concept described herein. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but the present invention may also include all embodiments and their equivalents falling within the scope of the present invention.
Claims (10)
Translated fromChinesePriority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610571691.XACN106203453B (en) | 2016-07-18 | 2016-07-18 | A kind of compressed sensing based biological and abiotic target identification method and its system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610571691.XACN106203453B (en) | 2016-07-18 | 2016-07-18 | A kind of compressed sensing based biological and abiotic target identification method and its system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106203453Atrue CN106203453A (en) | 2016-12-07 |
| CN106203453B CN106203453B (en) | 2019-05-28 |
Family
ID=57493621
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610571691.XAExpired - Fee RelatedCN106203453B (en) | 2016-07-18 | 2016-07-18 | A kind of compressed sensing based biological and abiotic target identification method and its system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106203453B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106595856A (en)* | 2017-02-13 | 2017-04-26 | 天津大学 | Terahertz wave compression perception rapid imaging apparatus based on rotation disc-like mask plate |
| CN106899305A (en)* | 2017-01-07 | 2017-06-27 | 陕西尚品信息科技有限公司 | A kind of primary signal reconstructing method based on Second Generation Wavelets |
| CN107273908A (en)* | 2017-07-06 | 2017-10-20 | 清华大学深圳研究生院 | A kind of target identification method based on compressed sensing |
| CN108280818A (en)* | 2018-01-19 | 2018-07-13 | 清华大学深圳研究生院 | A kind of fast target imaging method and system based on compressed sensing |
| CN113505695A (en)* | 2021-07-09 | 2021-10-15 | 上海工程技术大学 | AEHAL characteristic-based track fastener state detection method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102239793A (en)* | 2011-04-18 | 2011-11-16 | 浙江大学 | Real-time classification method and system of rice pests |
| CN103226196A (en)* | 2013-05-17 | 2013-07-31 | 重庆大学 | Radar target recognition method based on sparse feature |
| CN103398729A (en)* | 2013-07-31 | 2013-11-20 | 中国科学院空间科学与应用研究中心 | Compressed-sensing-based sparse aperture imaging system and method |
| CN103514456A (en)* | 2013-06-30 | 2014-01-15 | 安科智慧城市技术（中国）有限公司 | Image classification method and device based on compressed sensing multi-core learning |
| CN103996047A (en)* | 2014-03-04 | 2014-08-20 | 西安电子科技大学 | Hyperspectral image classification method based on compression spectrum clustering integration |
- 2016
- 2016-07-18CNCN201610571691.XApatent/CN106203453B/ennot_activeExpired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102239793A (en)* | 2011-04-18 | 2011-11-16 | 浙江大学 | Real-time classification method and system of rice pests |
| CN103226196A (en)* | 2013-05-17 | 2013-07-31 | 重庆大学 | Radar target recognition method based on sparse feature |
| CN103514456A (en)* | 2013-06-30 | 2014-01-15 | 安科智慧城市技术（中国）有限公司 | Image classification method and device based on compressed sensing multi-core learning |
| CN103398729A (en)* | 2013-07-31 | 2013-11-20 | 中国科学院空间科学与应用研究中心 | Compressed-sensing-based sparse aperture imaging system and method |
| CN103996047A (en)* | 2014-03-04 | 2014-08-20 | 西安电子科技大学 | Hyperspectral image classification method based on compression spectrum clustering integration |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106899305A (en)* | 2017-01-07 | 2017-06-27 | 陕西尚品信息科技有限公司 | A kind of primary signal reconstructing method based on Second Generation Wavelets |
| CN106899305B (en)* | 2017-01-07 | 2020-10-23 | 陕西尚品信息科技有限公司 | Original signal reconstruction method based on second-generation wavelet |
| CN106595856A (en)* | 2017-02-13 | 2017-04-26 | 天津大学 | Terahertz wave compression perception rapid imaging apparatus based on rotation disc-like mask plate |
| CN107273908A (en)* | 2017-07-06 | 2017-10-20 | 清华大学深圳研究生院 | A kind of target identification method based on compressed sensing |
| CN107273908B (en)* | 2017-07-06 | 2019-09-06 | 清华大学深圳研究生院 | A kind of compressed sensing based target identification method |
| CN108280818A (en)* | 2018-01-19 | 2018-07-13 | 清华大学深圳研究生院 | A kind of fast target imaging method and system based on compressed sensing |
| CN108280818B (en)* | 2018-01-19 | 2020-04-03 | 清华大学深圳研究生院 | Rapid target imaging method and system based on compressed sensing |
| CN113505695A (en)* | 2021-07-09 | 2021-10-15 | 上海工程技术大学 | AEHAL characteristic-based track fastener state detection method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106203453B (en) | 2019-05-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6616862B2 (en) | Image feature detection using edge vectors | |
| CN106845341B (en) | Unlicensed vehicle identification method based on virtual number plate | |
| Ma et al. | Fusioncount: Efficient crowd counting via multiscale feature fusion | |
| Khan et al. | Situation recognition using image moments and recurrent neural networks | |
| CN108564579B (en) | A kind of concrete crack detection method and detection device based on space-time correlation | |
| WO2016054779A1 (en) | Spatial pyramid pooling networks for image processing | |
| US20220383525A1 (en) | Method for depth estimation for a variable focus camera | |
| CN103440478B (en) | A kind of method for detecting human face based on HOG feature | |
| CN106203453A (en) | A kind of based on compressed sensing biological with abiotic target identification method and system thereof | |
| CN109214366A (en) | Localized target recognition methods, apparatus and system again | |
| CN110516731B (en) | Visual odometer feature point detection method and system based on deep learning | |
| CN112036381B (en) | Visual tracking method, video monitoring method and terminal equipment | |
| CN105005798B (en) | One kind is based on the similar matched target identification method of structures statistics in part | |
| CN104657717A (en) | Pedestrian detection method based on layered kernel sparse representation | |
| CN108665509A (en) | A kind of ultra-resolution ratio reconstructing method, device, equipment and readable storage medium storing program for executing | |
| CN114758145A (en) | Image desensitization method and device, electronic equipment and storage medium | |
| CN115272153A (en) | An image matching enhancement method based on feature sparse region detection | |
| Jiang et al. | Full-field deformation measurement of structural nodes based on panoramic camera and deep learning-based tracking method | |
| JP5704909B2 (en) | Attention area detection method, attention area detection apparatus, and program | |
| CN109508623A (en) | Item identification method and device based on image procossing | |
| Satish et al. | Hardware implementation of template matching algorithm and its performance evaluation | |
| CN110321869A (en) | Personnel's detection and extracting method based on Multiscale Fusion network | |
| KR101733288B1 (en) | Object Detecter Generation Method Using Direction Information, Object Detection Method and Apparatus using the same | |
| CN111951287A (en) | Two-dimensional code detection and recognition method | |
| Bhattacharya | HybridFaceMaskNet: A novel face-mask detection framework using hybrid approach |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee | Granted publication date:20190528 |