Target identification method for push-broom imaging of pyroelectric infrared column device

Abstract

The invention relates to the field of push-broom imaging of pyroelectric infrared detection, in particular to a target recognition method for push-broom imaging of a pyroelectric infrared column device. According to the method, the effective signal is judged based on the continuous slope change of each pixel in time, the problem that the amplitude of the output signal is influenced by the environment temperature is solved, the problem that unified threshold judgment is limited by the consistency difference of each pixel of a linear array device is also avoided, meanwhile, the influence of peak noise on target detection is prevented, and the application range is expanded; the proportion of the edge feature in the movement direction occupying the overall imaging range is taken as a key feature, and the problem that the detection and recognition accuracy is influenced along with the change of the target speed is solved; and meanwhile, the characteristic values required by calculation are reduced from N * channel number to 2 * channel number, so that the characteristic scale is reduced, and the recognition speed is further improved. Finally, the accuracy and speed of target recognition are improved, the problem that the existing recognition technology is limited by the target speed is solved, and the applicability is improved.

Description

Technical Field

[0001] The present invention relates to the field of push-scan imaging of pyroelectric infrared detection, and in particular to a target recognition method for push-scan imaging of pyroelectric infrared array devices, based on the extraction and recognition classification of edge features in the target motion direction. Background Art

[0002] Infrared imaging has a wide range of application needs in security, fire detection and other fields due to its all-weather detection capabilities that are not affected by lighting conditions. Among them, pyroelectric wire array devices occupy an important position in the field of infrared imaging detection due to their low power consumption, fast response and low cost.

[0003] However, the imaging mode of the pyroelectric infrared array device is push-scan imaging, and its imaging effect is affected by the scanning speed: first, the number and distribution of image data points generated by the same target after passing through the field of view of the linear array device at different speeds are different, and the area array imaging algorithm cannot be directly applied. In addition, too fast a push-scan speed may result in too few sampling points, which cannot fully reflect the shape characteristics of the target; while too slow a scanning speed will result in too many sampling points, causing a serious imbalance in the aspect ratio of the target and failing to accurately reflect the shape of the target.

[0004] Currently, most of the research on linear imaging recognition is aimed at imaging recognition of targets of a certain speed by intercepting images of a certain length. There are strict requirements on the speed of the target, and the recognition objects have great limitations. Summary of the invention

[0005] In view of the above-mentioned problems or shortcomings, in order to solve the problem of misjudgment of the recognition algorithm due to the change of imaging aspect ratio for the same target at different speeds in the push-scan imaging process of the existing pyroelectric infrared array device, the present invention provides a target recognition method for push-scan imaging of a pyroelectric infrared array device. Within the effective imaging range of the target, the number of sampling points of each pixel output before the effective response is generated and after the response is completely completed is counted, and divided by the number of sampling frames in the effective imaging range respectively, to obtain the proportion of the imaging time occupied by the non-effective signals on both sides of each pixel as the boundary feature of the target.

[0006] A target recognition method for push-scan imaging of a pyroelectric infrared array device comprises the following steps:

[0007] Step 1: The target with infrared radiation characteristics moves relative to the detection window of the pyroelectric infrared array device, and the output signal of each channel of the pyroelectric infrared array device is received in real time by the data acquisition board and transmitted to the host computer.

[0008] Step 2: Judge the signals of each channel obtained in step 1, determine the valid signal and binarize the data, where the valid signal is 1 and the invalid signal is 0; then splice the data of each channel in spatial order to obtain the binary target imaging, and cut the background to obtain the effective imaging range of the target.

[0009] Step 3, extract features from the binary target image obtained in step 2, and the extracted features are edge features in the target motion direction: within the effective imaging range of the target, count the number of sampling points of each pixel output before the effective response is generated and after the response is completely completed, and divide them by the number of sampling frames within the effective imaging range respectively; that is, the proportion of the imaging time occupied by the ineffective signals on both sides of each pixel is obtained as the edge feature of the target, so as to distinguish the time when different parts of the target enter and leave the field of view of each pixel.

[0010] Step 4: Repeat steps 1 to 3 to obtain no less than 100 groups of data and add the category to which each group of data belongs as the corresponding label, and use 60% to 90% of them as the training set and 10% to 40% as the test set.

[0011] Step 5: Use the training set obtained in step 4 as input data to establish a machine learning model, and use 3 to 8 times cross validation according to hyperparameter search to optimize the machine learning model; use the test set as input data to test the model recognition ability, and finally obtain a machine learning model with effective recognition ability.

[0012] Step 6: Collect and process data in real time through steps 1 to 3, and then connect the output data of step 3 to the machine learning model finally obtained in step 5 in real time to realize edge recognition and real-time classification of push-scan imaging of pyroelectric infrared array devices.

[0013] Furthermore, the valid signal judgment is based on the continuous change rate of the pixel signal: the change rate of the continuous response signal of a single pixel over time is calculated, and if the absolute value of the change rate between no less than four consecutive points is within the set threshold range (0.5V / s~50V / s), it is judged as a valid signal.

[0014] Furthermore, the imaging recognition is performed on a single target.

[0015] Furthermore, the effective imaging range means that within each effective imaging range, at least 2 pixels among all pixels in each sampling frame respond effectively, which is called a valid frame, and all valid frames must be continuous in time; if the valid frames are not continuous in time, they will be judged as multiple effective imaging ranges.

[0016] Furthermore, the machine learning model is a support vector machine.

[0017] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0018] 1. The effective signal judgment of the present invention is based on the continuous slope change of each pixel in time, which not only solves the problem that the output signal amplitude generated by the pyroelectric sensor for the same target is affected by the ambient temperature, but also avoids the problem that the unified threshold judgment is limited by the consistency difference of each pixel of the linear device and thus causes misjudgment. At the same time, it effectively prevents the influence of spike noise on target detection, and expands the application scope of the present imaging recognition method.

[0019] 2. The feature extraction method of the present invention selects the proportion of the edge feature in the motion direction to the overall imaging range as the key feature. Different parts of the target parallel to the detection window direction of the linear device radiate corresponding to different pixel numbers. Since the speed of each part of the target is the same in the motion direction, and the effective data length corresponding to different pixels changes proportionally with the speed, this feature is independent of the change in target speed. It effectively solves the problem that the imaging aspect ratio of the pyroelectric infrared push-scan imaging system changes with the target speed and thus affects the detection and recognition accuracy, and removes the restriction on target speed in the existing push-scan imaging recognition. At the same time, the characteristic values ​​required for calculation are also reduced from N×the number of channels to 2×the number of channels, where N is the number of sampling frames of the target effective imaging area. This feature extraction method effectively reduces the feature scale and thus improves the recognition calculation speed.

[0020] 3. In the present invention, by comparing and optimizing a variety of machine learning algorithms, a support vector machine is comprehensively selected to adapt to the current feature extraction method, which further improves the accuracy of target recognition and reduces the time consumed by target recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flow chart of the present invention.

[0022] Figure 2 Graph 1 shows a single effective imaging range and its binarization effect of an embodiment.

[0023] Figure 3 Confusion matrix of the recognition results for the example. DETAILED DESCRIPTION

[0024] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0025] A target recognition method for push-scan imaging of pyroelectric infrared array devices, such as Figure 1 As shown, the following steps are included:

[0026] Step 1: The gesture target with infrared radiation characteristics moves vertically relative to the detection window of the pyroelectric infrared array device. The output signal of each channel of the pyroelectric infrared array device is collected in real time by the acquisition board and transmitted to the host computer.

[0027] This embodiment uses a 32-element pyroelectric infrared array sensor, and the target with infrared radiation characteristics is the rock, scissors, and paper gestures. The target moves through the device field of view at different speeds in a direction perpendicular to the window of the array device, and the movement speed range is 0.1m / s to 2m / s. At the same time, the acquisition board collects the output signals of each channel of the pyroelectric infrared array device in real time and transmits them to the host computer.

[0028] Step 2: Judge the signals of each channel obtained in step 1, determine the valid signals and perform binarization on the data. Then, splice the data of each channel in spatial order and extract the effective imaging range of the target to obtain the binarized gesture image and the original image, as shown in Figure 2 shown.

[0029] Step 3: Extract edge features in the target motion direction from the binary target image obtained in step 2. Figure 2 In the binary gesture image shown on the right, the ineffective responses, i.e., black pixel blocks, on the left and right sides of each pixel are counted. If the pixel has no effective response throughout the entire process, the left and right sides are counted as N and 0, respectively, where N is the number of sampling frames in the effective imaging area. The count values ​​of each channel are then divided by the width N of the effective imaging area, and the ratio obtained is used as the feature value.

[0030] Step 4: To construct a data set for target recognition, repeat steps 1 to 3 to obtain data sets of three different gestures. Finally, 300 sets of data are collected and corresponding labels are added. 70% of them are used as training sets and 30% as test sets.

[0031] Step 5: Use the training set obtained in step 4 as input data to establish a machine learning model. Use 3-fold cross validation according to the hyperparameter search, iterate and optimize the machine learning model based on the results; use the test set as input data for testing, and finally obtain a machine learning model with a recognition rate of not less than 95%. The tested confusion matrix is ​​as follows: Figure 3 As shown in the figure, the effectiveness and superiority of this target recognition method are fully demonstrated.

[0032] Step 6. Finally, repeat steps 1 to 3, perform real-time feature extraction on the data output by the 32-element pyroelectric infrared array sensor and connect it to the support vector machine (machine learning model) obtained in step 5 to achieve edge recognition and real-time classification of push-scan imaging of the pyroelectric infrared array device. The system detection and recognition time reaches 30ms.

[0033] The specific implementation principle of the present invention is: during the push-scan imaging process, the aspect ratio of the target imaging will change proportionally with the push-scan speed, but the movement speed of each part of the target in the movement direction is the same, so the ratio of the edge features of the target in the movement direction to the overall image length will not change with the push-scan speed, thereby effectively solving the problem of image compression during the push-scan imaging process; at the same time, the eigenvalues ​​required to be calculated are also reduced from N×the number of channels to 2×the number of channels, where N is the number of sampling frames of the target imaging area. This recognition method effectively reduces the feature scale.

[0034] Through the above feature extraction method, although the feature extraction method based on the target motion direction characteristics is limited in the case of finely distinguishing targets or when the target edge features are not obvious, this feature extraction method can be well suited for target detection and identification in environments such as field detection, security alarm, etc., where the speed difference between different motion states of the same target is large, the real-time detection requirements are high, and the target edge features are obvious.

[0035] It can be seen from the above embodiments that the present invention first judges the effective signal based on the continuous slope change of each pixel in time, which not only solves the problem that the output signal amplitude is affected by the ambient temperature, but also avoids the problem that the unified threshold judgment is limited by the consistency difference of each pixel of the linear device, thereby causing misjudgment, and at the same time prevents the influence of spike noise on target detection, and expands the application scope of this imaging recognition method; secondly, the proportion of the edge feature in the motion direction occupying the overall imaging range is selected as the key feature, which is independent of the target speed change, and effectively solves the problem that the detection and recognition accuracy is affected by the change of the target speed; at the same time, the characteristic value required for calculation is also reduced from N× the number of channels to 2× the number of channels, which effectively reduces the feature scale, thereby reducing the pressure of the back-end algorithm, and then improves the recognition calculation speed; and the machine learning algorithm is assisted to reduce the recognition time while increasing the recognition accuracy. Finally, the present invention not only improves the accuracy and speed of target recognition, but also solves the problem that the existing recognition technology is limited by the target speed, and improves the applicability.

Claims

1. A target recognition method for push-scan imaging of a pyroelectric infrared array device, characterized in that: The steps include: Step 1: The target with infrared radiation characteristics moves relative to the detection window of the pyroelectric infrared array device, and the output signal of each channel of the pyroelectric infrared array device is received in real time by the data acquisition board and transmitted to the host computer; Step 2: judge the signals of each channel obtained in step 1, determine the valid signal and perform binarization processing on the data, where the valid signal is 1 and the invalid signal is 0; then splice the data of each channel in spatial order to obtain the binary target imaging, and cut the background to obtain the effective imaging range of the target; Step 3, extracting features from the binary target image obtained in step 2, the extracted features are edge features in the target motion direction: within the effective imaging range of the target, count the number of sampling points of each pixel output before the effective response is generated and after the response is completely completed, and divide them by the number of sampling frames within the effective imaging range respectively; Step 4: Repeat steps 1 to 3 to obtain no less than 100 sets of data and add the classification of each set of data as the corresponding label, and use 60% to 90% of them as the training set and 10% to 40% as the test set; Step 5: Use the training set obtained in step 4 as input data to establish a machine learning model, and use 3-8-fold cross validation to optimize the machine learning model according to hyperparameter search; use the test set as input data to test the recognition ability of the machine learning model, and finally obtain a machine learning model with effective recognition ability; Step 6: Collect and process data in real time through steps 1 to 3, and then connect the output data of step 3 to the final machine learning model obtained in step 5 in real time to achieve edge recognition and real-time classification for push-scan imaging of pyroelectric infrared array devices.

2. The target recognition method for push-scan imaging of a pyroelectric infrared array device as claimed in claim 1, characterized in that: The effective signal judgment is based on the continuous change rate of the pixel signal: the change rate of the continuous response signal of a single pixel over time is calculated. If the absolute value of the change rate between no less than four consecutive points is within the set threshold range of 0.5V / s to 50V / s, it is judged to be a valid signal.

3. The target recognition method for push-scan imaging of a pyroelectric infrared array device as claimed in claim 1, characterized in that: The imaging recognition is performed on a single target.

4. The target recognition method for push-scan imaging of a pyroelectric infrared array device as claimed in claim 1, characterized in that: The effective imaging range means that within each effective imaging range, at least two pixels among all the pixels in each sampling frame respond effectively, which is called a valid frame, and all valid frames must be continuous in time; if the valid frames are not continuous in time, they will be judged as multiple effective imaging ranges.

5. The target recognition method for push-scan imaging of a pyroelectric infrared array device as claimed in claim 1, characterized in that: The machine learning model is a support vector machine.

Images