A device damage detection method based on ultrasonic infrared thermal imaging time sequence self-weighting

Abstract

The application discloses a kind of based on ultrasonic infrared thermography timing self-weighting equipment damage detection method, comprising the following steps: one, the ultrasonic infrared thermography sequence of equipment to be measured is collected and is cropped to obtain infrared thermal map;Two, the superpixel block is obtained by using entropy rate superpixel segmentation method to infrared thermal map processing;Three, the pixel anchor point similarity matrix of each infrared thermal map is obtained;Four, based on pixel anchor point similarity matrix, set up fusion similarity matrix and the adaptive weighting weight of each infrared thermal map, construct ultrasonic infrared thermography timing self-weighting fusion model;Five, based on ultrasonic infrared thermography timing self-weighting fusion model optimization solution obtains optimal fusion similarity matrix;Six, based on optimal fusion similarity matrix using spectral clustering obtains damage region.The method step of the application is simple, based on fusion similarity matrix and realizes damage region segmentation, without artificial marking information, and dynamically adjust timing chart weight, realize unattended equipment damage detection.

Description

Technical Field

[0001] This invention belongs to the field of equipment damage image detection technology, specifically relating to a time-series self-weighted equipment damage detection method based on ultrasonic infrared thermal imaging. Background Technology

[0002] In non-destructive testing (NDT) techniques, traditional ultrasonic testing, while providing high-resolution information on internal defects, requires a coupling agent and is relatively slow. X-ray testing can penetrate materials and generate two-dimensional images of internal structures, but the equipment is expensive and poses radiation safety concerns. Ultrasonic infrared thermography, with its advantages of high speed, wide detection area, and strong adaptability, has long attracted widespread attention. This technology utilizes ultrasonic excitation to induce frictional heat generation in damaged areas within the inspected object, which can then be detected in real-time using an infrared thermal imager, offering unique advantages. Using ultrasonic infrared thermography for non-destructive testing of equipment effectively overcomes the limitations of existing traditional methods in the aforementioned aspects.

[0003] Accurately extracting the damaged area from ultrasound infrared thermography is a key issue. However, the blurred edges of damage caused by thermal diffusion in ultrasound infrared thermography can easily lead to misjudgments. Furthermore, due to differences in thermal diffusion between different devices, it is difficult to manually balance the specific time periods of heating, isothermal, and cooling processes, as well as the selection of input image frames. More importantly, current methods often require pre-labeling of the damaged area and rely on a large amount of labeled data for training.

[0004] Therefore, there is currently a lack of a well-designed device damage detection method based on temporal self-weighting of ultrasonic infrared thermal imaging, which can achieve adaptive weighting of ultrasonic infrared thermal imaging sequences, obtain the damage area by fusing similarity matrix, and realize unsupervised device damage detection without the need for manual annotation. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a device damage detection method based on time-series self-weighting of ultrasonic infrared thermal imaging, which addresses the shortcomings of the prior art. The method has simple steps and reasonable design, realizes adaptive weighting of ultrasonic infrared thermal imaging sequences, obtains the fused similarity matrix to obtain the damage area, and achieves unsupervised device damage detection without the need for manual annotation.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a device damage detection method based on time-series self-weighted ultrasonic infrared thermal imaging, the method comprising the following steps:

[0007] Step 1: Acquire the ultrasonic infrared thermal imaging sequence of the device under test and crop it to obtain an infrared thermal image;

[0008] Step 2: Use the entropy rate superpixel segmentation method to process the infrared thermal image to obtain superpixel blocks;

[0009] Step 3: Obtain the pixel anchor point similarity matrix for each infrared thermal image;

[0010] Step 4: Based on the pixel anchor similarity matrix, set the fusion similarity matrix and the adaptive weighting weight of each infrared thermogram to construct a time-series self-weighted fusion model for ultrasonic infrared thermal imaging;

[0011] Step 5: Optimize and solve the optimal fusion similarity matrix based on the temporal self-weighted fusion model of ultrasonic infrared thermal imaging;

[0012] Step 6: Based on the optimal fusion similarity matrix, use spectral clustering to obtain the damaged region.

[0013] The aforementioned time-series self-weighted equipment damage detection method based on ultrasonic infrared thermal imaging further includes the following step one:

[0014] Step 101: Use an ultrasonic infrared thermal imaging detection system to acquire infrared thermal images of the device under test to obtain an ultrasonic infrared thermal imaging sequence;

[0015] Step 102: Use a computer to extract the original infrared thermal image of frame t from the ultrasonic infrared thermal imaging sequence, and crop the edges of the original infrared thermal image of frame t to obtain the cropped infrared thermal image of frame t. The value of t ranges from 1 to N, and N represents the length of the ultrasonic infrared thermal imaging sequence.

[0016] Step 103: Repeat step 102 multiple times to complete the cropping of the original infrared thermal image of the Nth frame, and obtain the cropped infrared thermal image of the Nth frame.

[0017] Step 104: Record the infrared thermal images after cropping from frame 1 to frame N as N infrared thermal images.

[0018] The aforementioned time-series self-weighted equipment damage detection method based on ultrasonic infrared thermal imaging further includes the following step two:

[0019] Step 201: Combine the N infrared thermal images according to the number of channels to obtain a high-dimensional infrared thermal image of w×h×N; where w represents the width of the cropped infrared thermal image and h represents the height of the cropped infrared thermal image.

[0020] Step 202: Convert the w×h×N high-dimensional infrared thermal image into a two-dimensional matrix; where the number of rows in the two-dimensional matrix is ​​w×h and the number of columns in the two-dimensional matrix is ​​N;

[0021] Step 203: Use a computer to perform dimensionality reduction on the two-dimensional matrix using principal component analysis, and use the first principal component as the dimensionality-reduced two-dimensional matrix;

[0022] Step 204: Use a computer to convert the dimensionality-reduced two-dimensional matrix into a two-dimensional grayscale image;

[0023] Step 205: Using a computer, the entropy rate superpixel segmentation method is used to segment the two-dimensional grayscale image according to a set number of superpixel blocks I', resulting in I' superpixel blocks; where I' is a positive integer.

[0024] The aforementioned time-series self-weighted equipment damage detection method based on ultrasonic infrared thermal imaging further includes the following step three:

[0025] Step 301: Using a computer to... The anchor point of the i-th superpixel block on the t-th infrared thermal image is obtained. ;in, This represents the pixel value of the e-th pixel in the i-th superpixel block of the t-th infrared thermal image, where the value of e ranges from 0 to 1. The value of i ranges from 1 to I'. This represents the total number of pixels in the i-th superpixel block;

[0026] Step 302: Using a computer to... The anchor point between the j-th pixel and the i-th superpixel block on the t-th infrared thermal image is obtained. Similarity value between ;in, This represents the pixel value of the j-th pixel on the t-th infrared thermal image, where the value of j ranges from 1 to w×h;

[0027] Step 303: Repeat step 302 multiple times to obtain the similarity value between each pixel and each anchor point in the t-th infrared thermal image, and combine the similarity values ​​between each pixel and each anchor point to form the pixel-anchor-similarity matrix of the t-th infrared thermal image. .

[0028] The above-mentioned time-series self-weighted equipment damage detection method based on ultrasonic infrared thermal imaging further includes the following step four:

[0029] The computer sets the fusion similarity matrix corresponding to N infrared thermal images as follows: Input the pixel anchor point similarity matrix of the t-th infrared thermal image. Construct a time-series self-weighted fusion model of ultrasonic infrared thermal imaging Where min represents the minimum value and st represents the constraint condition. Denotes the square of the F-norm. Indicates hyperparameters, Represents the natural logarithm function. The number of rows is w×h, and the number of columns is I'; Let be the first constant matrix, and Each element in the array is 1, and Its size is I' × 1; It is the second constant matrix, and Each element in the array is 1, and The number of rows is w×h, and the number of columns is 1; This represents the adaptive weighting weight when fusing the t-th infrared thermal image.

[0030] The aforementioned method for equipment damage detection based on time-series self-weighted ultrasonic infrared thermal imaging further includes step five, which is as follows:

[0031] Step 501: Use a computer to initialize the adaptive weighting weights for fusing the various infrared thermal images to the following values: ;

[0032] Step 502: Input the pixel anchor point similarity matrix of the t-th infrared thermal image. and order Substituting the data into the time-series self-weighted fusion model of ultrasound infrared thermal imaging, we obtain the transformed time-series self-weighted fusion model of ultrasound infrared thermal imaging. ;

[0033] Step 503: Solve the converted ultrasonic infrared thermal imaging time-series self-weighted fusion model to obtain the optimal fusion similarity matrix after the first iteration. ;

[0034] Step 504: Using a computer to... The adaptive weighting weights for the t-th infrared thermal image fusion after the first iteration are obtained. ;

[0035] Step 505: Following the methods in steps 502 to 504, let The adaptive weighting weights are used when fusing the t-th infrared thermal image after the first iteration. Substituting the values ​​and solving the problem, we obtain the optimal fusion similarity matrix after the second iteration. Adaptive weighting weights when fusing with the t-th infrared thermal image after the second iteration ;

[0036] Step 506: Repeat step 505 multiple times, let The adaptive weighting weights used when fusing the t-th infrared thermal image after the (k-1)-th iteration Substituting the values ​​and solving the problem, we obtain the optimal fusion similarity matrix after the k-th iteration. Adaptive weighting weights when fusing with the t-th infrared thermal image after the k-th iteration Where k is a positive integer greater than 2;

[0037] Step 507: If the difference between the minimum value of the objective function in the k-th iteration and the minimum value of the objective function in the (k-1)-th iteration is less than 0.01, the iteration ends, and the optimal fusion similarity matrix after the k-th iteration is calculated. Let this be denoted as the optimal fusion similarity matrix.

[0038] The aforementioned method for equipment damage detection based on time-series self-weighted ultrasonic infrared thermal imaging further includes step six, which is as follows:

[0039] Step 601: Using a computer, the number of categories is set to 2. The optimal fusion similarity matrix is ​​clustered using the spectral clustering method to obtain pixels with category label 1 and pixels with category label 2.

[0040] Step 602: Use a computer to obtain the average pixel value of the pixel with category label 1 and the average pixel value of the pixel with category label 2. The pixel area corresponding to the category label with the larger average pixel value is the damaged area.

[0041] Compared with the prior art, the present invention has the following advantages:

[0042] 1. The method of this invention is simple and reasonably designed. First, an infrared thermal image is obtained by cropping the ultrasonic infrared thermal imaging sequence. Second, the infrared thermal image is processed by the entropy rate superpixel segmentation method to obtain superpixel blocks. Next, the pixel anchor point similarity matrix of each infrared thermal image is obtained. Then, an ultrasonic infrared thermal imaging time-series self-weighted fusion model is constructed to obtain the optimal fusion similarity matrix. Finally, the damage area is segmented based on the fusion similarity matrix classification. No manual annotation of information is required, and the weight of the time-series image is dynamically adjusted to obtain accurate damage areas in an unsupervised manner.

[0043] 2. This invention first uses the entropy rate superpixel segmentation method to process the infrared thermal image to obtain superpixel blocks, which makes it easier for the subsequent anchor point similarity matrix of each pixel to be based on a unified superpixel block, thus facilitating subsequent fusion.

[0044] 3. The present invention sets a fusion affinity matrix, which is obtained by fusing the pixel anchor point similarity matrix of each infrared thermogram. On the one hand, it reflects the damage detection of the ultrasonic infrared thermal imaging sequence; on the other hand, it can avoid the interference of unstable time series images to accurately reflect the stable area of ​​damage. In addition, it considers the adaptive weighting weight of each infrared thermogram, and the weight is updated adaptively to suppress the negative impact caused by the instability of early image frames in the heat diffusion stage. It avoids the subjective bias caused by the difference in heat diffusion under different damage conditions when manually extracting time series sequences, and effectively enhances the accuracy of damage segmentation.

[0045] 4. In the self-weighted fusion model of this invention, reasonable weights are generated by quantitatively differentiating the pixel anchor point similarity matrix of each infrared heatmap with the fusion results of all frames. The weights are further scaled by a designed function to increase the importance of stable frames and give unstable frames with large differences a scaled smaller weight, thereby improving robustness.

[0046] In summary, the method of this invention is simple in steps and reasonable in design. Based on the pixel anchor point similarity matrix and fusion similarity matrix of each infrared thermogram, and considering the adaptive weighting weight of each infrared thermogram, a time-series self-weighted fusion model of ultrasonic infrared thermal imaging is constructed to obtain the optimal fusion similarity matrix and achieve damage region segmentation. No manual annotation of information is required, and the weight of the time-series map is dynamically adjusted to achieve unsupervised equipment damage detection.

[0047] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0048] Figure 1 This is a flowchart of the method of the present invention.

[0049] Figure 2 This is a schematic diagram of damage detection after a 20J energy impact according to the present invention.

[0050] Figure 3 This is a schematic diagram of damage detection after a 40J energy impact according to the present invention. Detailed Implementation

[0051] like Figure 1 As shown, the present invention provides a time-series self-weighted device damage detection method based on ultrasonic infrared thermal imaging, which includes the following steps:

[0052] Step 1: Acquire the ultrasonic infrared thermal imaging sequence of the device under test and crop it to obtain an infrared thermal image;

[0053] Step 2: Use the entropy rate superpixel segmentation method to process the infrared thermal image to obtain superpixel blocks;

[0054] Step 3: Obtain the pixel anchor point similarity matrix for each infrared thermal image;

[0055] Step 4: Based on the pixel anchor similarity matrix, set the fusion similarity matrix and the adaptive weighting weight of each infrared thermogram to construct a time-series self-weighted fusion model for ultrasonic infrared thermal imaging;

[0056] Step 5: Optimize and solve the optimal fusion similarity matrix based on the temporal self-weighted fusion model of ultrasonic infrared thermal imaging;

[0057] Step 6: Based on the optimal fusion similarity matrix, use spectral clustering to obtain the damaged region.

[0058] In this embodiment, step one, the specific process is as follows:

[0059] Step 101: Use an ultrasonic infrared thermal imaging detection system to acquire infrared thermal images of the device under test to obtain an ultrasonic infrared thermal imaging sequence;

[0060] Step 102: Use a computer to extract the original infrared thermal image of frame t from the ultrasonic infrared thermal imaging sequence, and crop the edges of the original infrared thermal image of frame t to obtain the cropped infrared thermal image of frame t. The value of t ranges from 1 to N, and N represents the length of the ultrasonic infrared thermal imaging sequence.

[0061] Step 103: Repeat step 102 multiple times to complete the cropping of the original infrared thermal image of the Nth frame, and obtain the cropped infrared thermal image of the Nth frame.

[0062] Step 104: Record the infrared thermal images after cropping from frame 1 to frame N as N infrared thermal images.

[0063] In this embodiment, step two is as follows:

[0064] Step 201: Combine the N infrared thermal images according to the number of channels to obtain a high-dimensional infrared thermal image of w×h×N; where w represents the width of the cropped infrared thermal image and h represents the height of the cropped infrared thermal image.

[0065] Step 202: Convert the w×h×N high-dimensional infrared thermal image into a two-dimensional matrix; where the number of rows in the two-dimensional matrix is ​​w×h and the number of columns in the two-dimensional matrix is ​​N;

[0066] Step 203: Use a computer to perform dimensionality reduction on the two-dimensional matrix using principal component analysis, and use the first principal component as the dimensionality-reduced two-dimensional matrix;

[0067] Step 204: Use a computer to convert the dimensionality-reduced two-dimensional matrix into a two-dimensional grayscale image;

[0068] Step 205: Using a computer, the entropy rate superpixel segmentation method is used to segment the two-dimensional grayscale image according to a set number of superpixel blocks I', resulting in I' superpixel blocks; where I' is a positive integer.

[0069] In this embodiment, step three is as follows:

[0070] Step 301: Using a computer to... The anchor point of the i-th superpixel block on the t-th infrared thermal image is obtained. ;in, This represents the pixel value of the e-th pixel in the i-th superpixel block of the t-th infrared thermal image, where the value of e ranges from 0 to 1. The value of i ranges from 1 to I'. This represents the total number of pixels in the i-th superpixel block;

[0071] Step 302: Using a computer to... The anchor point between the j-th pixel and the i-th superpixel block on the t-th infrared thermal image is obtained. Similarity value between ;in, This represents the pixel value of the j-th pixel on the t-th infrared thermal image, where the value of j ranges from 1 to w×h;

[0072] Step 303: Repeat step 302 multiple times to obtain the similarity value between each pixel and each anchor point in the t-th infrared thermal image, and combine the similarity values ​​between each pixel and each anchor point to form the pixel-anchor-similarity matrix of the t-th infrared thermal image. .

[0073] In this embodiment, step four is as follows:

[0074] The computer sets the fusion similarity matrix corresponding to N infrared thermal images as follows: Input the pixel anchor point similarity matrix of the t-th infrared thermal image. Construct a time-series self-weighted fusion model of ultrasonic infrared thermal imaging Where min represents the minimum value and st represents the constraint condition. Denotes the square of the F-norm. Indicates hyperparameters, Represents the natural logarithm function. The number of rows is w×h, and the number of columns is I'; Let be the first constant matrix, and Each element in the array is 1, and Its size is I' × 1; It is the second constant matrix, and Each element in the array is 1, and The number of rows is w×h, and the number of columns is 1; This represents the adaptive weighting weight when fusing the t-th infrared thermal image.

[0075] In this embodiment, step five is as follows:

[0076] Step 501: Use a computer to initialize the adaptive weighting weights for fusing the various infrared thermal images to the following values: ;

[0077] Step 502: Input the pixel anchor point similarity matrix of the t-th infrared thermal image. and order Substituting the data into the time-series self-weighted fusion model of ultrasound infrared thermal imaging, we obtain the transformed time-series self-weighted fusion model of ultrasound infrared thermal imaging. ;

[0078] Step 503: Solve the converted ultrasonic infrared thermal imaging time-series self-weighted fusion model to obtain the optimal fusion similarity matrix after the first iteration. ;

[0079] Step 504: Using a computer to... The adaptive weighting weights for the t-th infrared thermal image fusion after the first iteration are obtained. ;

[0080] Step 505: Following the methods in steps 502 to 504, let The adaptive weighting weights are used when fusing the t-th infrared thermal image after the first iteration. Substituting the values ​​and solving the problem, we obtain the optimal fusion similarity matrix after the second iteration. Adaptive weighting weights when fusing with the t-th infrared thermal image after the second iteration ;

[0081] Step 506: Repeat step 505 multiple times, let The adaptive weighting weights used when fusing the t-th infrared thermal image after the (k-1)-th iteration Substituting the values ​​and solving the problem, we obtain the optimal fusion similarity matrix after the k-th iteration. Adaptive weighting weights when fusing with the t-th infrared thermal image after the k-th iteration Where k is a positive integer greater than 2;

[0082] Step 507: If the difference between the minimum value of the objective function in the k-th iteration and the minimum value of the objective function in the (k-1)-th iteration is less than 0.01, the iteration ends, and the optimal fusion similarity matrix after the k-th iteration is calculated. Let this be denoted as the optimal fusion similarity matrix.

[0083] In this embodiment, step six is ​​as follows:

[0084] Step 601: Using a computer, the number of categories is set to 2. The optimal fusion similarity matrix is ​​clustered using the spectral clustering method to obtain pixels with category label 1 and pixels with category label 2.

[0085] Step 602: Use a computer to obtain the average pixel value of the pixel with category label 1 and the average pixel value of the pixel with category label 2. The pixel area corresponding to the category label with the larger average pixel value is the damaged area.

[0086] In this embodiment, the device under test is specifically:

[0087] Carbon fiber reinforced polyether ether ketone composite material specimen plate, the length and height of the specimen plate are 150mm×100mm×3mm; the specimen plate is subjected to vertical impact tests with 20J and 40J energy respectively using a WanCe DI'T183E type drop hammer impact testing machine to obtain impact damage specimen plate.

[0088] In this embodiment, the impact-damaged sample is subjected to infrared thermal image acquisition using an ultrasonic infrared thermal imaging detection system.

[0089] In this embodiment, the ultrasonic infrared thermal imaging detection system includes an infrared thermal imager, an ultrasonic excitation source, and a computer. The infrared thermal imager is a VCHD research680 infrared thermal imager manufactured by InfraTec GmbH, Germany. The ultrasonic excitation source is a UST-200 ultrasonic gun manufactured by Nanjing Novell Optoelectronic Systems Co., Ltd. The acquired ultrasonic infrared thermal imaging sequence is sent to the computer for processing.

[0090] In this embodiment, N represents the length of the ultrasonic infrared thermal imaging sequence, and N=800;

[0091] The value of I' is 145.

[0092] In this embodiment, This indicates that each element in the matrix has a value greater than or equal to zero.

[0093] In this embodiment, it should be noted that the first principal component in step 203 is a one-dimensional vector with w×h rows and 1 column. By converting it into a w×h two-dimensional matrix, it becomes a two-dimensional matrix after dimensionality reduction.

[0094] In this embodiment, the original infrared thermal image in step 102 is a grayscale image with a size of 428×586. After cropping around the edges by no more than 1 / 3, and removing the edge effects of the ultrasonic excitation source and experimental device, the resulting image is a grayscale image with a size of 180×256, i.e., w is 180 and h is 256.

[0095] In this embodiment, during specific implementation, the hyperparameters The value range of is (0, 1). Different values ​​can be taken to make Based on the similarity matrix With fusion similarity matrix The difference in values ​​generates weights for different steepnesses. Further optimization is then performed. The value is 0.1, which can be determined according to... The difference values ​​are assigned to weighting coefficients with large differences to suppress the influence of the heat map on the heating and cooling processes.

[0096] In this embodiment, specifically in implementation, for the impact damage sample corresponding to 20J energy, such as... Figure 2In the middle (a), the ultrasound C-scan image is shown. (c)-(d) show the damage identification results obtained by processing the FastCD method (discrete spectral decomposition damage segmentation model), the WIASDMD method (wavelet sparse dictionary damage segmentation model), and the method of this invention, respectively. The average iteration times used by the FastCD method, the WIASDMD method, and the method of this invention are 14.35 seconds, 10.28 seconds, and 2.898 seconds, respectively. The ultrasound C-scan image is used as a benchmark for quantitative comparison, and the segmentation accuracies are 88.31%, 94.78%, and 96.72%, respectively.

[0097] For impact damage samples corresponding to 40J energy, such as Figure 3 In the middle (a), the ultrasound C-scan image is shown. (c)-(d) show the damage identification results obtained by processing the FastCD method (discrete spectral decomposition damage segmentation model), the WIASDMD method (wavelet sparse dictionary damage segmentation model), and the method of the present invention, respectively. The average iteration times used by the FastCD method, the WIASDMD method, and the method of the present invention are 18.31 seconds, 25.63 seconds, and 7.623 seconds, respectively, and the segmentation accuracies are 91.07%, 90.45%, and 97.58%, respectively.

[0098] In summary, the method of this invention is simple in steps and reasonable in design. Based on the pixel anchor point similarity matrix and fusion similarity matrix of each infrared thermogram, and considering the adaptive weighting weight of each infrared thermogram, a time-series self-weighted fusion model of ultrasonic infrared thermal imaging is constructed to obtain the optimal fusion similarity matrix and achieve damage region segmentation. No manual annotation of information is required, and the weight of the time-series map is dynamically adjusted to achieve unsupervised equipment damage detection.

[0099] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the present invention. Any simple modifications, alterations, or equivalent structural changes made to the above embodiments based on the technical essence of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A device damage detection method based on ultrasonic infrared thermography time series self-weighting, characterized by, The method comprises the following steps: Step one, collect the ultrasonic infrared thermal imaging sequence of the equipment to be tested and cut to obtain an infrared thermal image; Step two, use the entropy rate superpixel segmentation method to process the infrared thermal image to obtain a superpixel block; Step three, obtain a pixel anchor point similarity matrix of each infrared thermal image; Step four, based on the pixel anchor point similarity matrix, set a fusion similarity matrix and an adaptive weighting weight of each infrared thermal image, and construct an ultrasonic infrared thermal imaging time sequence self-weighted fusion model; Step five, based on the ultrasonic infrared thermal imaging time sequence self-weighted fusion model, the optimal fusion similarity matrix is obtained by optimization and solving; Step six, based on the optimal fusion similarity matrix, the damage area is obtained by using spectral clustering; Step two, the specific process is as follows: Step 201, combine N infrared thermal images according to the channel to obtain a high-dimensional infrared thermal image of w x h x N; wherein w represents the width of the cut infrared thermal image, and h represents the height of the cut infrared thermal image; Step 202, convert the high-dimensional infrared thermal image of w x h x N into a two-dimensional matrix; wherein the number of rows of the two-dimensional matrix is w x h, and the number of columns of the two-dimensional matrix is N; Step 203, use a computer to reduce the dimension of the two-dimensional matrix by principal component analysis, and take the first principal component as the two-dimensional matrix after dimension reduction; Step 204, use a computer to convert the two-dimensional matrix after dimension reduction into a two-dimensional gray image; Step 205, use a computer to segment the two-dimensional gray image according to the set number of superpixel blocks I' by using the entropy rate superpixel segmentation method, to obtain I' superpixel blocks; wherein I' is a positive integer; Step four, the specific process is as follows: The computer is used to set N infrared thermal images corresponding to the fusion similarity matrix as , input the pixel anchor point similarity matrix of the tth infrared thermal image , and construct an ultrasonic infrared thermal imaging time sequence self-weighted fusion model ; wherein min represents the minimum value, s.t. represents the constraint condition, represents the square of the F norm, represents the hyperparameter, represents the natural logarithm function, has w x h rows and I' columns; is a first constant matrix, and each element in the matrix is 1, and has I' x 1 size; is a second constant matrix, and each element in the matrix is 1, and has w x h rows and 1 column; represents the adaptive weighting weight during fusion of the tth infrared thermal image.

2. A method for detecting damage to equipment based on ultrasonic infrared thermography time series self-weighting according to claim 1, characterized in that: Step one, the specific process is as follows: Step 101, use an ultrasonic infrared thermal image detection system to collect infrared thermal images of the equipment to be tested to obtain an ultrasonic infrared thermal imaging sequence; Step 102, use a computer to take the t-th original infrared thermal image from the ultrasonic infrared thermal imaging sequence, and crop the edges of the t-th original infrared thermal image to obtain the t-th cropped infrared thermal image, wherein the value of t ranges from 1 to N, and N represents the length of the ultrasonic infrared thermal imaging sequence; Step 103, repeat step 102 multiple times to complete the cropping of the N-th original infrared thermal image to obtain the N-th cropped infrared thermal image; Step 104, the first cropped infrared thermal image to the N-th cropped infrared thermal image are denoted as N infrared thermal images.

3. A method for detecting damage in equipment based on ultrasonic infrared thermography time series self-weighting according to claim 1, characterized in that: Step three, the specific process is as follows: Step 301, using a computer to obtain the anchor point of the i-th superpixel block on the t-th infrared thermal image according to ; wherein, ; wherein, represents the pixel value of the e-th pixel point in the i-th superpixel block on the t-th infrared thermal image, the value range of e is , the value range of i is 1~I', represents the total number of pixel points in the i-th superpixel block. Step 302, using a computer to calculate the similarity value between the jth pixel point on the tth infrared thermal image and the anchor point of the ith superpixel block according to the following formula ; wherein, represents the pixel value of the jth pixel point on the tth infrared thermal image, and j has a value range of 1 to w x h.​​ Step 303, repeating step 302 multiple times to obtain the similarity value between each pixel point and each anchor point on the tth infrared thermal image, and combining the similarity value between each pixel point and each anchor point into a pixel-anchor similarity matrix of the tth infrared thermal image .

4. A method for detecting damage in equipment based on ultrasonic infrared thermography time series self-weighting according to claim 1, characterized in that: Step five, the specific process is as follows: Step 501, adopt computer to make each infrared thermal image fusion when the initial value of adaptive weighting weight is ; Step 502, input the pixel anchor point similarity matrix of the tth infrared thermal image Let Substitute the ultrasonic infrared thermal imaging time sequence self-weighted fusion model to obtain the converted ultrasonic infrared thermal imaging time sequence self-weighted fusion model ; Step 503, solving the converted ultrasonic infrared thermal imaging time sequence self-weighted fusion model to obtain an optimal fusion similarity matrix after first iteration ; Step 504, using a computer to obtain the adaptive weighting weight in the fusion of the tth infrared thermal image after the first iteration according to ;​ Step 505, according to the method of step 502 to step 504, let , the adaptive weighting weight of the tth infrared thermal image fusion at the first iteration , the optimal fusion similarity matrix after the second iteration is obtained by substituting and solving , and the adaptive weighting weight of the tth infrared thermal image fusion at the second iteration ; Step 506, repeating step 505 multiple times, let The adaptive weighting weight of the tth infrared thermal image fusion at the k-1th iteration is substituted and solved to obtain the optimal fusion similarity matrix after the kth iteration and the adaptive weighting weight of the tth infrared thermal image fusion at the kth iteration ; wherein k is a positive integer greater than 2; Step 507, the difference between the minimum value of the objective function of the kth iteration and the minimum value of the objective function of the k-1th iteration is less than 0.01, then the iteration ends, and the optimal fusion similarity matrix after the kth iteration is obtained is denoted as the optimal fusion similarity matrix.

5. A method for detecting damage in equipment based on ultrasonic infrared thermography time series self-weighting according to claim 4, characterized in that: Step six, the specific process is as follows: Step 601, use a computer to set the number of categories to 2, and use the spectral clustering method to cluster process the optimal fusion similarity matrix to obtain pixel points with a category label of 1 and pixel points with a category label of 2; Step 602, use a computer to obtain the pixel mean of the pixel points with a category label of 1 and the pixel mean of the pixel points with a category label of 2, and the pixel area corresponding to the category label with the larger pixel mean is the damage area.

Images