Thermal infrared image edge detection method and system based on eight-core edge extractor

Abstract

The invention provides a thermal infrared image edge detection method and system based on an eight-core edge extractor, and relates to the technical field of image processing and computer vision. According to the invention, the method comprises the steps: obtaining an initial image, carrying out the data preprocessing, carrying out the first wavelet transformation, carrying out the two-dimensional Gaussian kernel processing, inputting the processed image into a constructed eight-kernel edge extractor, obtaining an edge feature information graph, and carrying out the second discrete two-dimensional wavelet transformation, so as to further extract the multi-scale information of the edge features. The method effectively eliminates the influence of noise, improves the integrity of image information extraction, extracts the edge feature information of the image, and improves the accuracy and robustness of edge detection.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing and computer vision, and in particular, to a thermal infrared image edge detection method and system based on an eight-core edge extractor. Background Art

[0002] Edge detection is an important technology in image processing and computer vision, which is used to identify the set of pixel points with drastic brightness changes in an image, and extract the overall framework and structure of the image. As a very important feature, edges can help distinguish different objects and backgrounds. Correctly detecting image edges is crucial for analyzing image content, achieving image segmentation and positioning, etc. Edge detection can reduce the amount of image data, remove information irrelevant to the target, and retain the important structural attributes of the image.

[0003] With the development of science and technology, thermal infrared technology has been widely applied in various fields, including medical, military and other fields. Therefore, the edge detection of thermal infrared images has become a very important task. However, in reality, due to factors such as poor lighting conditions, improper camera parameter settings, environmental interference, and interference during data transmission in the process of processing, transmitting, collecting, and storing thermal infrared images, the image quality deteriorates, becomes blurred and mixed with noise. The addition of these random blurs and noises makes the edge detection of these degraded thermal infrared images a difficult problem.

[0004] Among the existing edge detection algorithms, they are mainly divided into first-order derivative algorithms and second-order derivative algorithms. However, both types of edge detection algorithms have weak edge detection capabilities, blurred edge extraction, inability to extract useful information, sensitivity to noise, detection of excessive noise, and the removal of useful edge information when removing noise in thermal infrared images. Currently, there are few algorithms specifically for edge detection of thermal infrared images. Therefore, in order to improve the edge detection effect of blurred and noise-containing thermal infrared images, an effective method that can suppress noise and accurately perform edge detection for thermal infrared images is urgently needed. Summary of the Invention

[0005] The main purpose of the present invention is to provide a thermal infrared image edge detection method and system based on an eight-core edge extractor, so as to solve the technical problems of noise sensitivity, blurred edge extraction, and serious lack of edge information existing in the processing of thermal infrared images in the prior art.

[0006] To solve the above technical problems, the technical solution adopted by the present invention is: a thermal infrared image edge detection method based on an eight-core edge extractor, including the following steps: Obtain an initial image, and after performing data preprocessing, perform a first wavelet transform; Perform two-dimensional Gaussian kernel processing on the image after the first-level wavelet transform; Construct an eight-core edge extractor, input the image processed by two-dimensional Gaussian kernel processing, and obtain an edge feature information map; Perform a second-level wavelet transform on the edge feature information map; Perform normalization processing on the image after the second-level wavelet transform to obtain a detection image.

[0007] In the preferred solution, the formula for the two-dimensional Gaussian kernel processing is: (8.1); In the formula, is the image before entering the eight-core edge extractor, is the two-dimensional array of the original image signal, that is, the pixel value matrix of the image, is the horizontal and vertical position coordinates of the pixel in the image, is the discrete two-dimensional wavelet transform, is the inverse discrete two-dimensional wavelet transform, is the soft threshold processing operation, is the two-dimensional Gaussian kernel function; The formula for obtaining the detection image is: (8.2); In the formula, N is the normalization processing, is the image obtained through the eight-core edge extractor, is the finally obtained edge detection result map.

[0008] In the preferred solution, the first-level wavelet transform and the second-level wavelet transform operations are the same, and specifically include the following steps: A1: Perform discrete two-dimensional wavelet transform on the picture to remove noise. The discrete two-dimensional wavelet transform is obtained by passing the picture through a high-pass filter and a low-pass filter. The formula is: (1.1); (1.2); In the formula, is the original image signal, are the horizontal and vertical position coordinates of the pixel in the image respectively, is the low-frequency subband coefficient; is the high-frequency subband coefficient, which is divided into three directions: is the horizontal detail, is the vertical detail, is the diagonal detail; is the scaling function, is the wavelet function, and j is the decomposition scale; M and N are the rows and columns of the image, and m and n are the indices of the spatial positions in the wavelet transform; A2: Perform soft threshold processing on the image after step A1. The formula is: (2.1); In the formula , is the current wavelet coefficient value, is the threshold, is the soft threshold function, is the sign function, used to retain the sign of x; A3: Finally, perform inverse two-dimensional discrete wavelet transform for image reconstruction. The formula is: (3.1); In the formula, is the overall contour or background information of the image, is the horizontal edge and horizontal detail information of the image, is the vertical edge and vertical detail information of the image, is the diagonal edge and diagonal detail information of the image, is the scale parameter, the row and column indices of the spatial position in the wavelet transform, are the column and row of the spatial coordinates of the image respectively.

[0009] In the preferred solution, the data preprocessing includes: performing grayscale processing on the acquired initial image, and converting it into a grayscale image by using the weighted average method, that is, for each pixel point, by performing weighted summation on the pixel values of the red (R), green (G), and blue (B) color channels of the image, where the weight of the green channel is 0.587, the weight of the red channel is 0.299, and the weight of the blue channel is 0.114. The formula is: (4.1); In the formula, is the grayscale value of the current pixel point, is the pixel value of the red channel, is the pixel value of the green channel, is the pixel value of the blue channel.

[0010] In the preferred solution, perform two-dimensional Gaussian kernel processing to make the image smoother, thereby suppressing noise. The formula is: (4.2); In the formula, are the standard deviations of the Gaussian kernel in the x and y directions respectively, are the means of the Gaussian kernel in the x and y directions respectively, is the value of the Gaussian kernel function, representing the intensity of the Gaussian kernel at the given position (x,y).

[0011] In the preferred solution, the construction of the eight-core edge extractor has the formula: (5.1); (5.2); In the formula, is the pixel point of the original image, and are the indexes used to represent the movement of the sub-kernel on the image, is to take the maximum value; is the edge intensity calculated in different directions, that is, the image gray value, where ; represents the final edge intensity; is the sub-kernel of the eight-core edge extractor, that is, the template coefficients in eight different directions.

[0012] In the preferred solution, the normalization process is performed to make the pixel value range of the image between the preset ranges and convert it into the preset format. The formula is: (6.1); In the formula, X is the original data value, Xmin is the minimum value in the dataset, and Xmax is the maximum value in the dataset.

[0013] In the preferred solution, a thermal infrared image edge detection system based on an eight-core edge extractor includes: An image acquisition module for acquiring an initial image, performing data preprocessing, and then performing the first wavelet transform; The first wavelet module for performing two-dimensional Gaussian kernel processing on the image after the first wavelet transform; A feature extraction module for constructing an eight-core edge extractor, inputting the image processed by the two-dimensional Gaussian kernel, and obtaining an edge feature information map; The second wavelet module for performing the second wavelet transform on the edge feature information map; A normalization module for performing normalization processing on the image after the second wavelet transform to obtain a detection image.

[0014] An electronic device includes a memory and a processor; The memory is used to store a computer program; The processor is used to implement the above-mentioned thermal infrared image edge detection method based on an eight-core edge extractor when executing the computer program.

[0015] A computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor, the described edge detection method for thermal infrared images based on an octa-core edge extractor is implemented.

[0016] The present invention provides an edge detection method for thermal infrared images based on an octa-core edge extractor. After obtaining an initial image and performing data preprocessing, a first-level wavelet transform is carried out, followed by two-dimensional Gaussian kernel processing. The processed image is input into the constructed octa-core edge extractor to obtain an edge feature information map. Then, a second-level discrete two-dimensional wavelet transform is performed to further extract the multi-scale information of the edge features. Finally, normalization processing is carried out to obtain the detection image, effectively eliminating the influence of noise, improving the integrity of image information extraction, extracting the edge feature information of the image, and improving the accuracy and robustness of edge detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention will be further described below in conjunction with the drawings and embodiments: Figure 1 It is the structural diagram of eight sub-cores in the octa-core edge extractor of the present invention; Figure 2 It is the flowchart of the edge detection method of the present invention; Figure 3 It is the comparison diagram of the effects of the present invention and the prior art; Figure 4 It is the flowchart of the edge detection algorithm Dwid of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] Embodiment 1 As Figures 1-4 shown, an edge detection method for thermal infrared images based on an octa-core edge extractor includes the following steps: S1: Obtain an initial image, and after performing data preprocessing, carry out a first-level wavelet transform.

[0019] S2: Perform two-dimensional Gaussian kernel processing on the image after the first-level wavelet transform. The formula is: (8.1); In the formula, is the image before entering the octa-core edge extractor, is the two-dimensional array of the original image signal, i.e., the image pixel value matrix, are the horizontal and vertical position coordinates of the pixel in the image respectively, is the discrete two-dimensional wavelet transform, is the inverse discrete two-dimensional wavelet transform, is the soft threshold processing operation, is the two-dimensional Gaussian kernel function.

[0020] S3: Construct an eight-core edge extractor, input the image processed in step S2, and obtain an edge feature information map.

[0021] S4: Perform a second-level wavelet transform on the edge feature information map.

[0022] S5: After normalizing the image after the second-level wavelet transform, obtain a detection image. The formula is: (8.2); In the formula, N is the normalization process, is the image obtained by the eight-core edge extractor, is the final edge detection result map obtained by the algorithm.

[0023] In this embodiment, by obtaining the initial thermal infrared image, performing necessary preprocessing on the initial image, performing discrete two-dimensional wavelet transform (DWT) on the preprocessed image, inputting the processed image into the constructed eight-core edge extractor to obtain an edge feature information map, and then performing a second-level discrete two-dimensional wavelet transform to further extract the multi-scale information of the edge features. Finally, through normalization processing to obtain a detection image, the edge feature information of the image is effectively extracted, and the accuracy and robustness of edge detection are improved.

[0024] In formula (8.1) of this embodiment, represents the original image signal, which is a two-dimensional array, that is, the image pixel value matrix; x is the column (or horizontal coordinate) in the image, starting from the left side of the image, usually starting from 0 and increasing; y is the row (or vertical coordinate) in the image, starting from the upper side of the image, usually starting from 0 and increasing.

[0025] In the preferred solution, the first-level wavelet transform and the second-level wavelet transform have the same operations, which specifically include the following steps: A1: Perform discrete two-dimensional wavelet transform on the picture to remove noise. The formula is: The discrete two-dimensional wavelet transform is obtained by passing the picture through a high-pass filter and a low-pass filter. The formula is: (1.1); (1.2); In the formula, is the original image signal, which is a two-dimensional array (i.e., the image pixel value matrix), is the horizontal and vertical position coordinates of the pixel in the image. is the low-frequency subband coefficient, representing the smooth part of the image (i.e., the low-frequency part, the contour or the overall information), is the high-frequency subband coefficient, which is divided into three directions: is the horizontal detail, For vertical details, For diagonal details, Is a scaling function used to extract low-frequency information (large-scale features), Is a wavelet function used to extract high-frequency information in different directions. j is the decomposition scale; M and N are the number of rows and columns of the image, i.e., the vertical and horizontal dimensions of the image, representing the size of the image, and m and n are the indices of the spatial positions in the wavelet transform respectively.

[0026] A2: Perform soft-thresholding on the image after step A1. The formula is: (2.1); In the formula, Is the current wavelet coefficient value, Represents the threshold, Is the soft-threshold function, Is the sign function used to retain the sign of x.

[0027] Is usually estimated from the noise level.

[0028] A3: Finally, perform the inverse discrete two-dimensional wavelet transform for image reconstruction. The formula is: (3.1); In the formula, Is the overall contour or background information of the image, Is the horizontal edge and horizontal detail information of the image, Is the vertical edge and vertical detail information of the image, Is the diagonal edge and diagonal detail information of the image; j is the scale parameter indicating the resolution level where the current wavelet coefficient is located. m and n represent the indices of the spatial positions in the wavelet transform, representing the specific positions of each component (low-frequency or high-frequency) in the image. x and y represent the spatial coordinates of the image, i.e., the positions of each pixel in the image. When j = 0, it represents the initial low-frequency component, representing the large-scale information (i.e., large structure and low-frequency part) of the image. As j increases, it represents finer scales and higher-frequency image features, such as the details of the image.

[0029] In this embodiment, the discrete two-dimensional wavelet transform is obtained by passing the image signal through a high-pass filter and a low-pass filter respectively to get the low-frequency subband coefficients and high-frequency subband coefficients. The high-frequency subband coefficients contain the edge and detail information of the image, while the low-frequency subband coefficients contain the overall contour or background information of the image.

[0030] In step A3, the inverse discrete two-dimensional wavelet transform uses the wavelet coefficients And To reconstruct the original image. The reconstruction process is divided into three stages: 1) In the first stage, the low-frequency sub-band coefficients are superimposed with the corresponding scaling function to restore the large contour information of the image.

[0031] 2) In the second stage, the high-frequency sub-band coefficients are superimposed with the corresponding wavelet function to restore the local details of the image.

[0032] 3) In the third stage, the frequency band information in four directions (low frequency + three high-frequency directions) is accumulated to finally obtain the complete image .

[0033] Through the above steps A1 - A3, first perform a discrete two-dimensional wavelet transform to remove noise, then use soft thresholding to remove the noise in the high-frequency sub-band coefficients while retaining the edge and detail information of the image. Finally, through the inverse discrete two-dimensional wavelet transform, image reconstruction is performed, and the processed low-frequency sub-band coefficients and high-frequency sub-band coefficients are recombined into an image signal.

[0034] In practical applications, the selection of the threshold λ can be adaptively adjusted according to the noise level and detail degree of the image.

[0035] In this embodiment, when performing a multi-scale wavelet transform, the above steps can be repeated multiple times, and the low-frequency sub-band coefficients are further decomposed each time to extract finer image features.

[0036] In a preferred solution, the data preprocessing includes: Performing grayscale processing on the obtained initial image, and using the weighted average method to convert it into a grayscale image to improve the accuracy and efficiency of subsequent processing.

[0037] Grayscale processing: For the obtained initial image, use the weighted average method to convert it into a grayscale image. In this process, for each pixel point, the pixel values of the three color channels of the image, namely red (R), green (G), and blue (B), are weighted and summed. The weight of the green channel is 0.587, the weight of the red channel is 0.299, and the weight of the blue channel is 0.114. This weighting strategy can highlight the difference in brightness information according to the human eye's perception sensitivity to different colors, thereby generating a grayscale image that conforms to visual perception. This processing process provides a clearer and more reasonable brightness distribution basis for subsequent image denoising, edge detection, and other processing steps. The specific formula is expressed as follows: (4.1); In the formula, is the grayscale value of this pixel point, is the pixel value of the red channel, is the pixel value of the green channel, is the pixel value of the blue channel.

[0038] In the preferred solution, two-dimensional Gaussian kernel processing is performed to make the image smoother, thereby suppressing noise. The formula is: (4.2); In the formula, are the standard deviations of the Gaussian kernel in the x-direction and y-direction respectively. These two parameters control the shape of the Gaussian kernel, that is, the degree of expansion in the horizontal and vertical directions. The larger the standard deviation, the wider the Gaussian kernel and the more obvious the smoothing effect. are the means (or centers) of the Gaussian kernel in the x-direction and y-direction respectively, is the value of the position Gaussian kernel function, representing the intensity of the Gaussian kernel at the given position (x, y).

[0039] In this embodiment, the two-dimensional Gaussian kernel is applied to the image to obtain a smoothed image. The convolution operation can be implemented by means of a sliding window, that is, the Gaussian kernel is weighted and summed with each pixel and its neighborhood in the image to obtain the smoothed pixel value, significantly reducing the noise in the image while retaining the edge and detail information of the image.

[0040] In the preferred solution, an eight-core edge extractor is constructed. The formula is: (5.1); (5.2); In the formula, is the pixel point of the original image, and are the indices used to represent the movement of the sub-kernel on the image, is to take the maximum value, is the edge intensity calculated in different directions, that is, the image gray value, where ; represents the final edge intensity; is the sub-kernel of the eight-core edge extractor, that is, the template coefficients in eight different directions, which are used to calculate the edge intensity in each direction.

[0041] In this embodiment, the sub-kernel of the eight-core edge extractor is used to calculate the image gray values of edge detection in eight different directions , and the gray values in the eight directions are taken as the maximum value as the final gray value of this pixel point. is the final edge intensity, and the maximum value is selected from the eight directions as the edge response of this pixel point.

[0042] In this embodiment, a thermal infrared image edge detection algorithm (Dwid) is constructed with an eight-core edge extractor, discrete two-dimensional wavelet transform, inverse discrete two-dimensional wavelet transform, soft threshold processing, and two-dimensional Gaussian kernel processing as the main body. The eight-core edge extractor is as shown in Figure 1 . The sub-kernels of the eight-core edge extractor are used to perform operations on each pixel point of the image and take the maximum value as the edge output of the image, so as to detect edges in different directions in the thermal infrared image. The sub-kernel is composed of 7.5, 0, -7.5, and each sub-kernel is responsible for detecting edges in different directions in the image. The sub-kernel is usually designed based on the Sobel operator, Prewitt operator or their variants to detect edges in a specific direction in the image.

[0043] Then, a second wavelet transform is performed to optimize the quality of the final result image. Finally, the test results of the thermal infrared image are compared with other advanced edge detection algorithms, as shown in Figure 3 .

[0044] As shown in Figure 4 , it is the flowchart of the overall edge detection algorithm (Dwid) of this application. Through these steps, the edge feature information map is subjected to discrete two-dimensional wavelet transform to remove the remaining noise, obtain more detailed information, and improve the image quality. After the wavelet transform, soft threshold processing and the corresponding inverse transform are performed, and the final image is obtained after normalization processing.

[0045] In the preferred solution, normalization processing is performed to make the pixel value range of the image between 0 and 255 and convert it into a preset format. The formula is: (6.1); In the formula, X is the original data value, Xmin is the minimum value in the data set, and Xmax is the maximum value in the data set.

[0046] In this embodiment, the preset range is set to 0 to 255. A new edge detection algorithm (Dwid) is designed based on the eight-core edge extractor. The overall flowchart of Dwid is as shown in Figure 4 . The sub-kernels of the eight-core edge extractor are used to perform operations on each pixel point of the image and take the maximum value as the edge output of the image, so as to detect edges in different directions in the thermal infrared image. Then, double wavelet transform, two-dimensional Gaussian kernel and soft threshold processing are added to retain the detailed features of the image and remove noise at the same time. Among them: the first wavelet transform is mainly used to perform preliminary denoising on the image and retain the main structure and edge information, and the second wavelet transform is used to optimize the quality of the final result image; finally, the test results of the thermal infrared image are compared with other advanced edge detection algorithms.

[0047] As shown in Figure 3As shown, compared with the images processed by seven algorithms such as Canny, Laplacian, DOG (Difference of Gaussians), Prewitt, Sobel, LOG (Laplacian of Gaussian), Scharr, Krisch, and Robert in the prior art, although Canny edge detection can extract some lines of vehicles and roads, the edges are severely missing, and the specific description is as follows: 1) Although the Laplacian and DOG operators can detect edges, due to the relatively dark brightness, the edge details are not obvious and the effect is weak; the Prewitt and Sobel operators can extract edge information better, but the details are missing and the edges are blurred; the Roberts operator is slightly stronger than the Laplacian and DOG operators, but the extracted edges are still weak and the brightness is on the dark side; although LOG edge detection can detect certain edges, due to the too low brightness, many details are missing; the Scharr and Kirsch operators can extract edges relatively clearly, however, they are still greatly affected by the noise in the thermal infrared image, resulting in noise being misjudged as edges.

[0048] 2) From the comparison results, it can be seen that the Dwid operator in this embodiment performs best in complex and degraded thermal infrared images, can clearly extract edge information, effectively suppress noise at the same time, extracts edge information to the greatest extent, shows the best edge detection effect, and greatly improves the edge extraction clarity in thermal infrared image processing.

[0049] The technical solution of this embodiment solves the technical problems of noise sensitivity, fuzzy edge extraction, and serious edge information loss existing in the edge detection algorithm in the prior art in thermal infrared images, more effectively eliminates the influence of noise in thermal infrared images, and improves the clarity of image edges.

[0050] Embodiment 2 Further described in combination with Embodiment 1, a thermal infrared image edge detection system based on an eight-core edge extractor is provided, including: An image acquisition module, used to acquire an initial image, and after performing data preprocessing, perform a first-level wavelet transform; A first-level wavelet module, used to perform two-dimensional Gaussian kernel processing on the image after the first-level wavelet transform; A feature extraction module, used to construct an eight-core edge extractor, input the image processed by two-dimensional Gaussian kernel processing, and obtain an edge feature information map; A second-level wavelet module, used to perform a second-level wavelet transform on the edge feature information map; A normalization module, used to perform normalization processing on the image after the second-level wavelet transform to obtain a detection image.

[0051] An electronic device, comprising a memory and a processor; The memory is used for storing a computer program; The processor is used for implementing the thermal infrared image edge detection method based on an octa-core edge extractor as in Embodiment 1 when executing the computer program.

[0052] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the thermal infrared image edge detection method based on an octa-core edge extractor as in Embodiment 1 is implemented.

[0053] The above embodiments are only the preferred technical solutions of the present invention and should not be regarded as limitations on the present invention. The protection scope of the present invention should be the technical solutions recorded in the claims, including the equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, the equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A thermal infrared image edge detection method based on an eight-core edge extractor, characterized in that: The following steps are involved: Obtain the initial image, perform data preprocessing, and then perform the first wavelet transform; The image after the first wavelet transformation is processed with a two-dimensional Gaussian kernel; Construct an eight-core edge extractor, input the image processed by two-dimensional Gaussian kernel, and obtain the edge feature information map; Perform a second wavelet transform on the edge feature information graph; The image after the second wavelet transform is normalized to obtain the detection image.

2. The thermal infrared image edge detection method based on the eight-core edge extractor according to claim 1 is characterized in that: The two-dimensional Gaussian kernel processing formula is: （8.1）； In the formula, This is the image before entering the eight-core edge extractor. is the two-dimensional array of the original image signal, i.e. the pixel value matrix of the image. is the horizontal and vertical coordinates of the pixel in the image, is the discrete two-dimensional wavelet transform, is the inverse discrete two-dimensional wavelet transform, is a soft thresholding operation, is a two-dimensional Gaussian kernel function; The formula for obtaining the detection image is: （8.2）； In the formula, N is the normalization processing, is the image obtained by the eight-core edge extractor, This is the final edge detection result image.

3. The thermal infrared image edge detection method based on the eight-core edge extractor according to claim 1 is characterized in that: The first wavelet transform and the second wavelet transform have the same operation, specifically including the following steps: A1: Perform discrete two-dimensional wavelet transform on the image to remove noise. Discrete two-dimensional wavelet transform is obtained by passing the image through a high-pass filter and a low-pass filter. The formula is: （1.1）； （1.2）； In the formula, is the original image signal, are the horizontal and vertical coordinates of the pixel in the image, respectively. is the low frequency subband coefficient; is the high frequency subband coefficient, which is divided into three directions: For horizontal details, For vertical details, For diagonal details; is the scale function, is the wavelet function, j is the decomposition scale; M and N are the rows and columns of the image, and m and n are the indexes of the spatial position in the wavelet transform; A2: Perform soft threshold processing on the image after step A1. The formula is: 　　　　　　（2.1）； In the formula, is the current wavelet coefficient value, is the threshold value, is the soft threshold function, is a sign function, used to preserve the positive and negative signs of x; A3: Finally, perform inverse discrete two-dimensional wavelet transform to reconstruct the image. The formula is: （3.1）； In the formula, is the overall outline or background information of the image, is the horizontal edge and horizontal detail information of the image, is the vertical edge and vertical detail information of the image, is the diagonal edge and diagonal detail information of the image, is the scale parameter, The row and column indices of the spatial locations in the wavelet transform, are the columns and rows of the image space coordinates, respectively.

4. The thermal infrared image edge detection method based on the eight-core edge extractor according to claim 1 is characterized in that: The data preprocessing includes: graying the acquired initial image and converting it into a grayscale image by using a weighted average method, that is, for each pixel point, weighted summing is performed on the pixel values ​​of the three color channels of the image, namely, the weight of the green channel is 0.587, the weight of the red channel is 0.299, and the weight of the blue channel is 0.114, and the formula is: （4.1）； In the formula, is the gray value of the current pixel, is the pixel value of the red channel, is the pixel value of the green channel, is the pixel value of the blue channel.

5. The thermal infrared image edge detection method based on the eight-core edge extractor according to claim 1 is characterized in that: The two-dimensional Gaussian kernel processing is performed to make the image smoother, thereby suppressing noise. The formula is: （4.2）； In the formula, are the standard deviations of the Gaussian kernel in the x and y directions, are the means of the Gaussian kernel in the x and y directions, respectively. is the value of the Gaussian kernel function, which represents the strength of the Gaussian kernel at a given position (x, y).

6. The thermal infrared image edge detection method based on the eight-core edge extractor according to claim 1 is characterized in that: The formula for constructing the eight-core edge extractor is: 　　　　　　（5.1）； （5.2）； In the formula, is the pixel of the original image, and is the index used to represent the movement of the sub-core on the image, To obtain the maximum value, is the edge strength calculated in different directions, that is, the image grayscale value, where ; To represent the final edge strength; It is the sub-core of the eight-core edge extractor, that is, the template coefficients in eight different directions.

7. The thermal infrared image edge detection method based on the eight-core edge extractor according to claim 1 is characterized in that: The normalization process is performed so that the pixel value range of the image is within a preset range, and the image is converted into a preset format, and the formula is: 　　　　　（6.1）； Where X is the original data value, Xmin is the minimum value in the data set, and Xmax is the maximum value in the data set.

8. A thermal infrared image edge detection system based on an eight-core edge extractor, characterized in that: include: An image acquisition module is used to acquire an initial image, perform data preprocessing, and then perform a first wavelet transform; The first wavelet module is used to perform two-dimensional Gaussian kernel processing on the image after the first wavelet transformation; The feature extraction module is used to construct an eight-core edge extractor, input the image processed by the two-dimensional Gaussian kernel, and obtain the edge feature information map; The second wavelet module is used to perform a second wavelet transform on the edge feature information graph; The normalization module is used to obtain a detection image after normalizing the image after the second wavelet transform.

9. An electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is used to implement the thermal infrared image edge detection method based on the eight-core edge extractor as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the thermal infrared image edge detection method based on the eight-core edge extractor as described in any one of claims 1 to 7 is implemented.

Images