A circuit board abnormal heating component positioning method based on thermal images

Abstract

This invention discloses a method for locating abnormally overheating components on circuit boards based on thermal images. Starting from the characteristics of the thermal image itself, it obtains difference images under normal and abnormal conditions, and performs image registration using an accelerated robust feature registration algorithm optimized based on the Laplacian operator. This ensures that the difference images under abnormal and normal conditions are well aligned after registration, resulting in a clear differential thermal image. Then, outlier detection is performed on the differential thermal image using the Laplacian distribution, obtaining a differential thermal image after outlier detection, thus initially identifying abnormally overheating regions and locating the abnormally overheating components. Finally, kernel density estimation is used to determine the number of K-means clusters, and the K-means algorithm is used to segment the differential thermal image after outlier detection, obtaining a K-means segmented image. Morphological opening-closing operations are then performed on the image to remove minute lines. The thermal image segmentation results of this invention can accurately provide information on abnormally overheating components, enabling accurate location and identification of these components, helping to narrow down the defect range and providing defect judgment information.

Description

Technical Field

[0001] This invention belongs to the field of circuit board fault detection technology, and more specifically, relates to a method for locating abnormally heated components on a circuit board based on thermal images. Background Technology

[0002] Circuit boards are indispensable components in electronic devices, serving to physically support and connect electronic elements. However, circuit boards are prone to mechanical and soldering failures during production and use, causing electronic products to malfunction. Therefore, fault detection is necessary. Traditional circuit board fault detection typically employs contact-based measurement methods, requiring inspectors to use measuring instruments such as multimeters and voltmeters to measure specific circuit parameters and compare them with standard circuit data to determine if a fault has occurred. However, this measurement method is not only cumbersome and has a limited scope for each inspection, but also requires inspectors to be highly familiar with the principles of the circuit board being tested and its PCB layout.

[0003] Emerging non-contact fault detection methods mainly include machine vision based on optical images, X-ray inspection, and terahertz detection. However, these methods primarily detect physical defects in circuit boards, such as delamination, voids, and internal cracks. For functional faults in circuit boards, non-contact detection methods based on infrared thermal imaging (thermal images) are employed. These methods offer advantages such as safety, efficiency, wide-area inspection, and intuitive defect display, making them suitable for detecting various functional faults in circuit boards, such as open circuits, short circuits, and parameter deviations. Functional faults in circuit boards directly or indirectly affect their logic structure, causing changes in the voltage or current across components. This affects the heat output of the components, altering the overall heat distribution trend after the circuit board is powered on. Detecting these thermal changes allows for the location of abnormal components.

[0004] The localization of abnormally overheating components relies on the comparison between thermal images of standard circuit boards and faulty circuit boards, depending on image registration and segmentation. However, due to the characteristics of thermal images, such as blurred edges and lack of high-frequency texture information, image registration and segmentation techniques fail in the thermal imaging field, making it difficult to locate abnormally overheating components. Therefore, thermal image registration and segmentation remain the current challenges in locating abnormally overheating components, and how to better perform thermal image registration and segmentation is a problem that needs to be solved. Summary of the Invention

[0005] The purpose of this invention is to address the difficulty in locating abnormally heated components caused by failures in thermal imaging in existing image registration and image segmentation technologies. This invention provides a method for locating abnormally heated components on circuit boards based on thermal images, so as to achieve more accurate location of abnormally heated elements and regions.

[0006] To achieve the aforementioned objectives, this invention proposes a Speeded Up Robust Features (SURF) algorithm based on Laplacian operator optimization and a K-means algorithm based on Laplacian distribution improvement to process thermal images of circuit boards under normal and abnormal conditions, thereby enabling the localization of abnormally heated components on the circuit board. Specifically:

[0007] The present invention provides a method for locating abnormally heated components on a circuit board based on thermal imaging, characterized by comprising the following steps:

[0008] (1) Use an infrared thermal imager to capture thermal images of the circuit board under test in its normal state before and after power-on. , Thermal images under abnormal conditions , ;

[0009] (2) Obtain the difference image:

[0010]

[0011]

[0012] in, This is the difference image under normal conditions. This is a difference image under abnormal conditions;

[0013] (3) An accelerated robust feature registration algorithm based on Laplacian operator optimization is adopted for the difference image under normal conditions. Difference images under abnormal conditions Image registration was performed to obtain the difference image under abnormal conditions. Difference image under normal conditions spatial transformation matrix ;

[0014] (4) Difference images under abnormal conditions Based on the spatial transformation matrix Perform the transformation to obtain the difference image compared to the normal state. Difference image under anomalous alignment The difference between the two is then calculated to obtain a differential thermal image. :

[0015]

[0016] (5) Use the K-means algorithm based on the Laplace distribution to process the differential thermal image. Image segmentation

[0017] 5.1) Using Laplace distribution on differential thermal images Outlier detection is performed to obtain differential thermal images after outlier detection.

[0018] First, calculate the median of the pixel values. and mean absolute deviation :

[0019]

[0020]

[0021] in, It is a differential thermal image The The grayscale value of each pixel For differential thermal images The number of pixels;

[0022] Then, determine the low probability threshold. and high threshold , respectively as probabilities Substitute the following into the inverse cumulative distribution function of the Laplace distribution:

[0023]

[0024] The obtained grayscale value The corresponding image low grayscale thresholds Image high grayscale threshold ;

[0025] Finally, remove the differential thermal image. The medium gray value is located at the low gray value threshold of the image. Image high grayscale threshold The difference thermal image after outlier detection is obtained by analyzing the pixels between them. ;

[0026] 5.2) Determining the number of K-means clusters using kernel density estimation

[0027] Differential thermal image after outlier detection Kernel density estimation is performed, resulting in a kernel density estimation curve:

[0028]

[0029] in, This is an estimate of the probability density function. Differential thermal image The first in The grayscale value of each pixel Differential thermal image The number of pixels, Represents bandwidth. Represents the Gaussian kernel function;

[0030] Among them, bandwidth Determined by Silverman's Law:

[0031]

[0032] in, This represents the sample variance.

[0033] The number of main peaks in the kernel density curve is detected, and this number of main peaks is the number of K-means clusters.

[0034] 5.3) Based on the K-means cluster number, use the K-means algorithm to process the differential thermal image after outlier detection. Perform segmentation to obtain the K-means segmented image. ;

[0035] 5.4) For images A morphological opening-closing operation is performed to remove minute lines, resulting in the final segmentation result that locates the abnormal heating element and region.

[0036] The objective of this invention is achieved as follows:

[0037] This invention relates to a method for locating abnormally overheating components on circuit boards based on thermal images. It starts from the characteristics of the thermal images themselves and obtains the difference image under normal conditions. Difference images under abnormal conditions Furthermore, an accelerated robust feature registration algorithm based on Laplacian operator optimization is employed for image registration, resulting in a difference image under abnormal conditions after registration. Difference image with normal state It can be well aligned, resulting in a clear differential thermal image. Then, the Laplace distribution was used to analyze the differential thermal image. Perform outlier detection and obtain differential thermal images after outlier detection. Thus, the differential thermal image was initially determined. The abnormal heating areas were identified, and the abnormal heating elements were located. Finally, kernel density estimation was used to determine the number of K-means clusters, and the K-means algorithm was used to analyze the differential thermal image after outlier detection. Perform segmentation to obtain the K-means segmented image. For images Morphological opening-closing operations are performed to remove minute lines. The thermal image segmentation results of this invention can accurately provide information on abnormally overheating components, enabling accurate location and identification of these components, thus helping to narrow down the defect range and provide defect judgment information. Attached Figure Description

[0038] Figure 1 This is a flowchart of a specific implementation of the method for locating abnormally heated components on a circuit board based on thermal images according to the present invention;

[0039] Figure 2 These are images of two circuit boards used in the experiment, where (a) is experimental circuit board A and (b) is experimental circuit board B.

[0040] Figure 3 These are preprocessed images of experimental circuit board A in this embodiment, where (a) is a steady-state thermal image before power-on, (b) is a steady-state image after power-on, and (c) is a difference image;

[0041] Figure 4 These are the Laplacian edge detection results in this embodiment, where (a) represents the normal state of circuit board A, (b) represents the fault state of circuit board A, (c) represents the normal state of circuit board B, and (d) represents the fault state of circuit board B.

[0042] Figure 5 This is a difference image between the normal state and the abnormal state of the circuit board after registration in this embodiment. Among them, (a) is the normal state of circuit board A, (b) is the fault state 1 of circuit board A, (c) is the fault state 2 of circuit board A, (d) is the fault state 3 of circuit board A, (e) is the fault state 4 of circuit board A; (f) is the normal state of circuit board B, (g) is the fault state 1 of circuit board B, and (h) is the fault state 2 of circuit board B.

[0043] Figure 6 This is a comparison between the fitted Laplace distribution PDF and the difference image histogram in this embodiment, where (a) is experimental circuit board A and (b) is experimental circuit board B;

[0044] Figure 7 These are the outlier detection images and their grayscale distribution histograms in this embodiment, where (a) is the outlier detection image and (b) is the grayscale distribution histogram.

[0045] Figure 8These are the kernel density estimation curves of the difference images in this embodiment, where (a) represents normal experimental circuit board A, (b) represents fault state 1 of experimental circuit board A, (c) represents fault state 2 of experimental circuit board A, (d) represents fault state 3 of experimental circuit board A, (e) represents fault state 4 of experimental circuit board A, (f) represents normal experimental circuit board B, (g) represents fault state 1 of experimental circuit board B, and (h) represents fault state 2 of experimental circuit board B.

[0046] Figure 9 These are the final image segmentation results in this embodiment, where (a) represents normal experimental circuit board A, (b) represents fault state 1 of experimental circuit board A, (c) represents fault state 2 of experimental circuit board A, (d) represents fault state 3 of experimental circuit board A, (e) represents fault state 4 of experimental circuit board A, (f) represents normal experimental circuit board B, (g) represents fault state 1 of experimental circuit board B, and (h) represents fault state 2 of experimental circuit board B. Detailed Implementation

[0047] The specific embodiments of the present invention will now be described with reference to the accompanying drawings to enable those skilled in the art to better understand the invention. It should be particularly noted that in the following description, detailed descriptions of known functions and designs that might obscure the main content of the invention will be omitted here.

[0048] Figure 1 This is a flowchart of a specific implementation of the method for locating abnormally heated components on a circuit board based on thermal images according to the present invention.

[0049] like Figure 1 As shown, this invention acquires thermal images of standard and faulty circuit boards and preprocesses them. Then, edge detection and Gaussian filtering are applied to the thermal images. Finally, SURF image registration is used to register the images of the standard and faulty circuit boards. After registration, differential thermal images are obtained. Outlier detection and kernel density estimation are performed, followed by K-means clustering for image segmentation. Morphological processing is then applied to locate abnormal heating elements and regions. Specifically, in this embodiment, as... Figure 1 As shown, the method for locating abnormally heated components on a circuit board based on thermal images according to the present invention includes the following steps:

[0050] Step S1: Acquire thermal images of the circuit board under normal and fault conditions.

[0051] Use an infrared thermal imager to capture thermal images of the circuit board under test in its normal state before and after power-on. , Thermal images under abnormal conditions , (raw format file). The resulting data format is shown in equation (1).

[0052] (1)

[0053] in, Indicates the location at pixel coordinates ( , The infrared radiation intensity of the pixel at position ) For the vertical spatial resolution of infrared thermal imagers, This represents the horizontal spatial resolution of the thermal imager.

[0054] In this embodiment, data is collected as follows: Figure 2 The thermal images of experimental circuit board A and experimental circuit board B before power-on and after power-on reaching thermal equilibrium are shown. The obtained thermal images can be represented by equation (1).

[0055] Step S2: Preprocess the acquired thermal image

[0056] The difference between the final thermal equilibrium temperature of a component and the ambient temperature is positively correlated with the component's own heat generation power. Therefore, to ensure that the thermal image accurately reflects the changes in the heat generation power of the circuit board components, and to suppress background noise in the thermal image, areas where the temperature change before and after heating is insignificant are removed from the thermal image of the circuit board after power-on and reaching thermal equilibrium. Experimentally, thermal images of the circuit board before and after power-on were collected, and the difference was calculated to obtain the difference image.

[0057] (2)

[0058] (3)

[0059] in, This is the difference image under normal conditions. This is a difference image under abnormal conditions.

[0060] The difference image of the obtained infrared thermal image , This is positively correlated with the heat dissipation power of the circuit board components. In this embodiment, Figure 3 The images show the difference between the experimental circuit board A before and after power-on, as well as before and after power-on.

[0061] Step S3: Process the obtained difference image , SURF image registration with Laplacian operator optimization

[0062] An accelerated robust feature registration algorithm based on Laplacian operator optimization is adopted for difference images under normal conditions. Difference images under abnormal conditions Image registration was performed to obtain the difference image under abnormal conditions. Difference image under normal conditions spatial transformation matrix .

[0063] Since thermal images lack high-frequency information, Laplacian edge detection is performed first, followed by SURF feature point detection, to obtain the difference image under abnormal conditions. Difference image under normal conditions spatial transformation matrix In this embodiment, the specific steps include:

[0064] Step S3.1: Use the Laplacian operator to extract the edges of the preprocessed difference image.

[0065] Extracting the difference image under normal conditions using the Laplacian operator. Difference images under abnormal conditions The edges are obtained to create an edge image. , In this embodiment, the Laplacian operator uses a 4-neighborhood Laplacian template for edge image extraction, as shown in equation (4).

[0066] (4)

[0067] Using the 3×3, 4-neighborhood convolution template of the Laplacian operator, i.e., the 4-neighborhood Laplacian template (see equation (4)), the difference image is convolved to obtain the following: Figure 4 The Laplacian edge detection results are shown.

[0068] Step S3.2: Use the SURF registration method to process the edge image extracted by the Laplacian operator. , Perform Gaussian filtering: Use a Gaussian filter on the edge image , Gaussian filtering is performed to obtain the filtered edge image. , The Gaussian filter function is shown in equation (5).

[0069] (5)

[0070] in, denoted as the scale of the Gaussian filter function.

[0071] Step S3.3: Use the SURF image registration algorithm to process the filtered edge image. , Perform image registration and calculate edge images. To the edge image spatial transformation matrix That is, the difference image under abnormal conditions. Difference image under normal conditions spatial transformation matrix In this embodiment, the specific steps include:

[0072] 1. By adjusting the scale of the second-order Gaussian differential operator The size of the convolution template is used to convolve the difference image to generate a multi-scale response map, thus constructing a scale pyramid for the image.

[0073] 2. Calculate the target pixel and its 3x3 matrix using the determinant of the approximate Hessian matrix. 3 The response values ​​of the target pixel are compared with those of the other 26 pixels in the 3-neighborhood (including spatial and scale neighborhoods). If the target pixel is a local feature point and its response value is greater than the threshold, it is retained as a local feature point of the region. The determinant of the approximate Hessian matrix is ​​shown in equation (6).

[0074] (6)

[0075] in, The determinant of the approximate Hessian matrix. , , These are the corresponding box filter and the response value after image convolution, respectively. The weighting coefficients are used to compensate for the errors caused by replacing the Gaussian convolution template with a box filter.

[0076] 3. Obtained local feature points The precision is at the pixel level. To obtain sub-pixel level local feature points, the Taylor expansion (retaining the second-order terms) is used to expand the response function (see Equation (6)) at the feature points and the derivative is zero to obtain the feature point offset. See equation (7).

[0077] (7)

[0078] in is the coordinate vector of the feature point in scale space.

[0079] 4. Determine a principal direction for each feature point; at the feature point... Within the neighborhood, take the feature point as the center and the angle as 60°. The fan-shaped region is rotated around the center. The horizontal and vertical Haar wavelet responses within the window are calculated. The horizontal response is recorded when the sum of the horizontal and vertical responses reaches its maximum value. and vertical response , main direction It can be calculated using equation (15).

[0080] (8)

[0081] 5. Extract feature descriptors along the principal direction of the feature points; along the principal direction, take a region of size 20 centered on the feature points. 20 A rectangular region was defined, and this region was divided into 16 sub-blocks. The Haar wavelet response of the sampling points within each sub-block was calculated using a scale of 3.3. The Gaussian function is used to weight the sum of the level responses of all sampling points within the sub-block. Sum of vertical responses The sum of the absolute values ​​of the horizontal responses and the sum of the absolute values ​​of the vertical response The eigenvectors are obtained as shown in equation (16). By concatenating the feature vectors of the 16 sub-blocks, we obtain a 64-dimensional feature descriptor for a single feature point.

[0082] (9)

[0083] 6. Perform coarse matching on points with similar feature point descriptors between the two images. Feature point matching depends on the Euclidean distance between feature point descriptors. Calculate the Euclidean distance between the feature point descriptors of the reference image and the feature point descriptors of the image to be registered. Find the minimum and second minimum distances between the feature point descriptors of the image to be registered and the reference image, and calculate the ratio between the minimum and the second minimum distances. If the ratio is less than the threshold of 0.6, retain the feature point pair.

[0084] 7. Use the M-Estimator Sample Consensus (MSAC) algorithm to perform fine matching of feature points; the MSAC algorithm is essentially minimizing the function shown in Equation (17).

[0085] (10)

[0086] in, This refers to the reprojection error between feature point pairs. The MSAC loss kernel function can be represented by equation (11).

[0087] (11)

[0088] in, The threshold value is set.

[0089] 8. Using the least squares method, calculate the spatial transformation matrix of the two images based on the matching point pairs obtained. And spatial transformation is performed on the thermal images of the fault.

[0090] Step S4: Extract the difference image under abnormal conditions. Based on the spatial transformation matrix Perform the transformation to obtain the difference image compared to the normal state. Difference image under anomalous alignment The difference between the two is then calculated to obtain a differential thermal image. :

[0091] (12)

[0092] Figure 5 This is a differential thermal image of the registered circuit board in the normal and abnormal states in this embodiment. .Depend on Figure 5 It can be seen that the thermal image registration method proposed in this invention can complete the thermal image registration of different circuit boards and different working states of the circuit boards, and can obtain the differential thermal image between the registered fault thermal image and the normal thermal image by relying on the image registration method. .

[0093] Step S5: Use the K-means algorithm based on the Laplace distribution to process the differential thermal image. Image segmentation

[0094] Step S5.1: Apply Laplace distribution to the differential thermal image Outlier detection is performed to obtain differential thermal images after outlier detection.

[0095] The probability density function of the Laplace distribution is shown in equation (13):

[0096] (13)

[0097] in, The location parameter determines the peak position of the Laplace distribution. The scale parameter determines the degree of dispersion of the distribution. According to maximum likelihood estimation, The median of the pixel values. This represents the mean absolute deviation.

[0098] First, calculate the median of the pixel values. and mean absolute deviation :

[0099] (14)

[0100] (15)

[0101] in, It is a differential thermal image The The grayscale value of each pixel For differential thermal images The number of pixels.

[0102] Then, determine the low probability threshold. and high threshold , respectively as probabilities Substitute the following into the inverse cumulative distribution function of the Laplace distribution:

[0103] (16)

[0104] The obtained grayscale value The corresponding image low grayscale thresholds Image high grayscale threshold .

[0105] Finally, remove the differential thermal image. The medium gray value is located at the low gray value threshold of the image. Image high grayscale threshold The difference thermal image after outlier detection is obtained by analyzing the pixels between them. .

[0106] Figure 6 This example compares the fitted Laplace distribution PDF with the histogram of the difference image. Figure 6 It can be seen that, overall, the gray value distribution of the difference image follows the same trend as the Laplace distribution. Therefore, the Laplace distribution is used to model the difference thermal image, retaining values ​​with a probability less than 0.01 in the Laplace distribution.

[0107] Figure 7 This is the difference thermal image and its grayscale distribution histogram after outlier detection in this embodiment. Figure 7 (a) It can be seen that after outlier detection, abnormal heating areas in the differential thermal image can be preliminarily identified, and abnormal heating elements can be located. Figure 7 (b) It can be seen that the gray-level histogram of the image after outlier detection generally shows a multi-peak distribution, which indicates that the data after outlier detection is closer to a clustered distribution and is more suitable for processing by the K-means algorithm.

[0108] Step S5.2: Determine the number of K-means clusters using kernel density estimation.

[0109] Thermal images after outlier detection Kernel density estimation is performed to obtain the kernel density estimation curve.

[0110] Differential thermal image after outlier detection Kernel density estimation is performed, resulting in a kernel density estimation curve:

[0111] (17)

[0112] in, This is an estimate of the probability density function. Differential thermal image The first in The grayscale value of each pixel Differential thermal image The number of pixels, Represents bandwidth. Represents the Gaussian kernel function

[0113] Among them, bandwidth Determined by Silverman's Law:

[0114] (18)

[0115] in This represents the sample variance.

[0116] Kernel density estimation is used to estimate an unknown probability density function from a finite sample. It detects the number of dominant peaks in the kernel density curve, which represents the number of K-means clusters. The number of dominant peaks in the kernel density estimation curve can represent the number of clusters in the difference thermal image after outlier detection, thus determining the number of clusters in the K-means algorithm. The number of dominant peaks in the kernel density curve can be calculated using the following algorithm:

[0117] 1. Set the minimum peak spacing and significance threshold for the main peak;

[0118] 2. The locations of all peaks are initially determined by calculating the maximum points of the curve;

[0119] 3. Significance of each maximum point: Traverse leftward from the peak position, find the first point higher than the peak (or reach the starting point), and record the lowest point in the range from the peak to the left until a point higher than the peak (or reach the starting point). Similarly, traverse rightward from the peak position, find the first point higher than the peak (or reach the ending point), and record the lowest point in the range from the peak to the right until a point higher than the peak (or reach the starting point). The significance is calculated as the peak height minus the height of the higher of the two lowest values.

[0120] 4. Remove the maxima with significance less than the threshold to obtain the position and height of the main peak of the kernel density estimation;

[0121] 5. Remove all other extreme points within the minimum peak spacing of a main peak extreme point.

[0122] Figure 8 This is the kernel density estimation curve of the difference image in this embodiment. Figure 8 It can be seen that all kernel density estimation curves exhibit at least one or more peaks. If there are components with heating power greater than normal power, a peak will form to the right of the mean of the differential thermal image; if there are components with heating power less than normal power, a peak will form to the left of the mean of the differential thermal image. Therefore, the number of main peaks in the kernel density estimation curve can represent the number of clusters in the differential thermal image after outlier detection, and this can be used to calculate the number of K-means clusters.

[0123] Step S5.3: Based on the K-means cluster count, use the K-means algorithm to process the differential thermal image after outlier detection. Perform segmentation to obtain the K-means segmented image. .

[0124] The main steps of the K-means algorithm are as follows:

[0125] 1. Determine the number of clusters in K-means clustering. Initial selection One initial cluster center;

[0126] 2. Calculate the Euclidean distance from each sample point to each cluster center, and assign the sample point to the cluster corresponding to the nearest cluster center;

[0127] 3. After each sample point is divided, calculate the new cluster centers, which are the average values ​​of the data points within each cluster;

[0128] 4. If the cluster centers converge or the maximum number of iterations is reached, output the clustering results; otherwise, update the cluster centers and repeat steps (2) to (3).

[0129] Step S5.4: Process the image A morphological opening-closing operation is performed to remove minute lines, resulting in the final segmentation result that locates the abnormal heating element and region.

[0130] Figure 9 This is the final image segmentation result in this embodiment. The final thermal image segmentation result obtained by the method in this paper can accurately provide information on abnormally heated components, realize the location and identification of abnormally heated components, help narrow down the defect range, and provide defect judgment information.

[0131] Although the illustrative specific embodiments of the present invention have been described above to enable those skilled in the art to understand the invention, it should be understood that the invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the invention as defined and determined by the appended claims, and all inventions utilizing the concept of the present invention are protected.

Claims

1. A method for locating abnormally overheating components on a circuit board based on thermal images, characterized in that, Includes the following steps: (1) Use an infrared thermal imager to capture thermal images of the circuit board under test in its normal state before and after power-on. , Thermal images under abnormal conditions , ; (2) Obtain the difference image: ； ； in, This is the difference image under normal conditions. This is a difference image under abnormal conditions; (3) An accelerated robust feature registration algorithm based on Laplacian operator optimization is adopted for the difference image under normal conditions. Difference images under abnormal conditions Image registration was performed to obtain the difference image under abnormal conditions. Difference image under normal conditions spatial transformation matrix ; (4) Difference images under abnormal conditions Based on the spatial transformation matrix Perform the transformation to obtain the difference image compared to the normal state. Difference image under anomalous alignment The difference between the two is then calculated to obtain a differential thermal image. : ； (5) Use the K-means algorithm based on the Laplace distribution to process the differential thermal image. Perform image segmentation; 5.1) Using Laplace distribution on differential thermal images Outlier detection is performed to obtain differential thermal images after outlier detection. ; First, calculate the median of the pixel values. and mean absolute deviation : ； ； in, It is a differential thermal image The The grayscale value of each pixel For differential thermal images The number of pixels; Then, determine the low probability threshold. and high threshold , respectively as probabilities Substitute the following into the inverse cumulative distribution function of the Laplace distribution: ； The obtained grayscale value The corresponding image low grayscale thresholds Image high grayscale threshold ; Finally, remove the differential thermal image. The medium gray value is located at the low gray value threshold of the image. Image high grayscale threshold The difference thermal image after outlier detection is obtained by analyzing the pixels between them. ; 5.2) Use kernel density estimation to determine the number of K-means clusters; Differential thermal image after outlier detection Kernel density estimation is performed, resulting in a kernel density estimation curve: ； in, This is an estimate of the probability density function. Differential thermal image The first in The grayscale value of each pixel Differential thermal image The number of pixels, Represents bandwidth. Represents the Gaussian kernel function; Among them, bandwidth Determined by Silverman's Law: ； in, This represents the sample variance. The number of main peaks in the kernel density curve is detected, and this number of main peaks is the number of K-means clusters. 5.3) Based on the K-means cluster number, use the K-means algorithm to process the differential thermal image after outlier detection. Perform segmentation to obtain the K-means segmented image. ; 5.4) For images A morphological opening-closing operation is performed to remove minute lines, resulting in the final segmentation result that locates the abnormal heating element and region.

2. The method for locating abnormally heated components on a circuit board based on thermal images according to claim 1, characterized in that, Step (3) describes the use of an accelerated robust feature registration algorithm based on Laplacian operator optimization for the difference image under normal conditions. Difference images under abnormal conditions Image registration was performed to obtain the difference image under abnormal conditions. Difference image under normal conditions spatial transformation matrix for: 3.1) Extracting the difference image under normal conditions using the Laplacian operator. Difference images under abnormal conditions The edges are obtained to create an edge image. , ; 3.2) Using the SURF registration method to extract edge images from the Laplacian operator , Perform Gaussian filtering: Use a Gaussian filter on the edge image , Gaussian filtering is performed to obtain the filtered edge image. , ; 3.3) Use the SURF image registration algorithm to process the filtered edge image. , Perform image registration and calculate edge images. To the edge image spatial transformation matrix That is, the difference image under abnormal conditions. Difference image under normal conditions spatial transformation matrix .

Images