Battery film surface defect detection method based on deep learning

Abstract

The invention relates to the technical field of defect detection, and provides a battery film surface defect detection method based on deep learning, and the detection method comprises the following steps: S1, obtaining the image features of a battery film surface image, the image features comprising a color feature and an HSV feature; s2, extracting a battery film surface image according to a V-channel image in the HSV features, and removing an irrelevant background to obtain a first image; s3, performing non-local mean filtering and logarithm transformation on the first image in sequence to obtain a second image; s4, performing connectivity analysis on the second image to screen out defect areas and extract defect features; s5, by analyzing defect areas and features, positioning edge positions of defect features by using an edge model; and S6, detecting the edge position of the defect feature in the step S5 by adopting a sequential similarity detection algorithm, and matching the defect target of the surface image of the primary battery film. The objective of the invention is to improve the accuracy and efficiency of defect detection of the perovskite cell film.

Description

Technical Field

[0001] The present invention relates to the technical field of defect detection, and in particular to a method for detecting surface defects of battery thin films based on deep learning. Background Art

[0002] In recent years, energy and environmental issues have received more and more attention. Perovskite solar cells have been widely studied due to their low cost and high efficiency. Although the energy conversion efficiency of perovskite solar cells has made great progress, various surface defects often occur during the experimental preparation process, limiting their quality and performance. Unlike large-scale solar panels, photovoltaic films prepared in the laboratory have the characteristics of small area, smooth surface and easy reflection.

[0003] Traditional detection methods such as manual detection, mechanical contact detection and traditional visual detection have problems such as low sampling rate, low accuracy, poor real-time performance and low efficiency. Therefore, it is difficult to detect small spots, scratches, missing corners and other defects on the surface of small photovoltaic films prepared in experiments using traditional defect detection methods, which is far from meeting the current experimental requirements. At present, there are few detection methods for surface defects of small-area perovskite solar cell films prepared in the laboratory. Relying on traditional manual detection methods is time-consuming, laborious and inefficient, and is far from meeting the requirements of industrial production.

[0004] Therefore, improving the accuracy and speed of surface defect detection of perovskite solar cells is a crucial part of the industrial production process. Summary of the invention

[0005] In view of the above defects, the purpose of the present invention is to propose a method for detecting surface defects of battery thin films based on deep learning, aiming to improve the accuracy and efficiency of defect detection of perovskite battery thin films.

[0006] To achieve this object, the present invention adopts the following technical solutions:

[0007] A method for detecting surface defects of a battery thin film based on deep learning, the method comprising the following steps:

[0008] S1: Acquire image features of the battery film surface image, including color features and HSV features;

[0009] S2: extracting the battery film surface image according to the light and dark difference of the V channel image in the HSV feature, removing irrelevant background, and obtaining the first image;

[0010] S3: performing non-local mean filtering and logarithmic transformation on the first image in sequence to obtain a second image;

[0011] S4: performing connectivity analysis on the second image to screen out defect areas and extract defect features;

[0012] S5: By analyzing the defect area and features, the edge position of the defect feature is located using the edge model;

[0013] S6: Using a sequential similarity detection algorithm to detect the edge position of the defect feature in step S5, matching the defect target of the original battery thin film surface image.

[0014] Preferably, performing connectivity analysis on the second image to screen out defect areas and extract defect features includes:

[0015] According to the specific starting node and ending node of the enhanced image, whether the two points are connected is analyzed, so as to find a single part and extract the foreground part;

[0016] Through the feature information, the threshold is set to filter out the defect area and extract the defect features.

[0017] Preferably, the edge position of the defect feature in step S5 is detected using a sequential similarity detection algorithm, and the defect target matching the surface image of the original battery film includes:

[0018] Make a matching template according to the edge position of the defect feature;

[0019] A sequential similarity detection algorithm is used to scan the template sub-images corresponding to the pixels in the original image one by one to match all defect targets.

[0020] Preferably, obtaining the battery film surface image includes:

[0021] Acquire the RGB three-channel color image of the battery film surface image, and perform normalization processing to obtain the color characteristics of the battery film surface image;

[0022] The normalization process satisfies the relationship: R′=R / 255, G′=G / 255, B′=B / 255;

[0023] Among them, R represents the red channel, G represents the red channel, B represents the red channel, and R′, G′, and B′ refer to the normalized values ​​of R, G, and B, respectively.

[0024] Furthermore, the HSV features of the battery film surface image are obtained including:

[0025] Calculate the values ​​of H, S and V to satisfy the relationship:

[0026] V = max(R′, G′, B′)

[0027]

[0028]

[0029] If H is less than 0, add 360 to H;

[0030] Convert H, S and V to between 0 and 255 to obtain HSV features, and satisfy the relationship:

[0031] Preferably, the non-local mean filtering satisfies the relationship:

[0032]

[0033] Among them, I(i, j) represents the pixel value of the surface image of the original battery film, Ω represents the search range of the image, and w(x, y, i, j) represents the weight calculated according to the similarity of the image blocks, and satisfies the relationship:

[0034]

[0035] Among them, I(Ω(x, y)) and I(Ω(i, j)) represent the image blocks centered at (x, y) and (i, j), respectively, and h represents the parameter controlling the weight attenuation.

[0036] Furthermore, the logarithmic transformation satisfies the relationship:

[0037]

[0038] in, represents the image pixel value after non-local mean filtering, represents the pixel value of the image after logarithmic transformation, and c represents a constant value.

[0039] Preferably, the edge model satisfies the relationship:

[0040]

[0041] Among them, M(x, y; A, B, σ, u, v) represents the intensity at the corresponding pixel position, (x, y) represents a pixel position in the image, A represents the intensity of the image background, B represents the peak intensity in the bright area, (u, v) represents the position of the peak in the bright area, and σ 2 represents the Gaussian model variance.

[0042] One of the above technical solutions has the following advantages or beneficial effects:

[0043] The present invention uses RGB and HSV models to extract features from images, which can fully capture color information. By separating hue, saturation and brightness, the robustness of color information under different lighting conditions is enhanced, which helps to more accurately identify the state and potential defects of battery film. Non-local mean filtering can effectively remove noise from the image while retaining edge information to avoid detail blurring that may be caused by traditional mean filtering. Combined with logarithmic transformation, the overall contrast of the battery film image is improved, making the dark details of the battery film more obvious. Through connectivity analysis, the target is converted from pixel level to connected component level, which helps to identify and screen out continuous defective areas and further improve the accuracy of defect detection. The edge model can accurately identify the location of defect features, thereby improving positioning accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0045] Figure 1 is a flow chart of a method for detecting surface defects of a battery thin film based on deep learning provided in an embodiment of the present invention;

[0046] Figure 2 is a schematic diagram of an image acquisition platform of a battery thin film surface defect detection method based on deep learning provided in an embodiment of the present invention;

[0047] Figure 3 is an image of a defective state before being processed by the battery thin film surface defect detection method based on deep learning provided by an embodiment of the present invention;

[0048] Figure 4 is an image of a normal state before being processed by the battery thin film surface defect detection method based on deep learning provided by an embodiment of the present invention;

[0049] Figure 5 is a first detection result diagram of a battery thin film surface defect detection method based on deep learning provided by an embodiment of the present invention;

[0050] Figure 6 is a second detection result diagram of the battery thin film surface defect detection method based on deep learning provided by an embodiment of the present invention;

[0051] Figure 7 It is an edge image coordinate system of a battery thin film surface defect detection method based on deep learning provided by an embodiment of the present invention;

[0052] Figure 8 It is a sub-pixel edge schematic diagram of a battery thin film surface defect detection method based on deep learning provided by an embodiment of the present invention;

[0053] Among them, there are a device bracket 1, an image acquisition device 2, a ring light source 3, a sample to be tested 4, a storage table 5, a rotating base 6 of a rotatable storage table, a cylindrical bracket 7, and a device base 8. DETAILED DESCRIPTION

[0054] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0055] In the present invention, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of more restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.

[0056] Traditional detection methods such as manual detection, mechanical contact detection and traditional visual detection have problems such as low sampling rate, low accuracy, poor real-time performance and low efficiency. Therefore, it is difficult to detect small spots, scratches, missing corners and other defects on the surface of small photovoltaic films prepared in experiments using traditional defect detection methods, which is far from meeting the current experimental requirements. At present, there are few detection methods for surface defects of small-area perovskite solar cell films prepared in the laboratory. Relying on traditional manual detection methods is time-consuming, laborious and inefficient, and is far from meeting the requirements of industrial production.

[0057] Therefore, a method for detecting surface defects of battery thin films based on deep learning is proposed. The detection method can be used in combination with an image acquisition platform, such as Figure 2As shown, the image acquisition platform is composed of an image acquisition device 2, an annular light source 3, a rotating base 6 of a rotatable storage platform, a cylindrical bracket 7, an equipment base 8, a storage table 5 and an equipment bracket 1. Among them, the image acquisition device adopts an MV-E series industrial camera, which has the characteristics of ultra-high resolution, high precision, high definition, good color reproduction, low noise, etc. In addition, considering the reflection of the glaze surface of square mosaic ceramic tiles, the service life and stability of the light source, the commonly used annular light source, backlight source or strip light source cannot achieve a good effect, so the light source needs to adopt the MV-RC-V series coaxial light source, which emits light from the side and has a special semi-transparent and semi-reflective lens structure, uniform brightness, stable performance, suitable for surface defect detection of easily reflective objects, and has a good application prospect in industrial production detection. During image acquisition, the relative distance between the lens position of the image acquisition device 2 and the sample to be tested 4 is kept fixed, and the circular rotatable platform is composed of the storage table 5 and the rotating base 6 of the rotatable storage table, which can easily and quickly rotate the sample to be tested 4 to obtain images at different angles.

[0058] like Figure 1 As shown, in a preferred embodiment of the present invention, the detection method comprises the following steps:

[0059] S1: Acquire image features of the battery film surface image, including color features and HSV features;

[0060] S2: extracting the battery film surface image according to the light and dark difference of the V channel image in the HSV feature, removing irrelevant background, and obtaining the first image;

[0061] S3: performing non-local mean filtering and logarithmic transformation on the first image in sequence to obtain a second image;

[0062] S4: performing connectivity analysis on the second image to screen out defect areas and extract defect features;

[0063] S5: By analyzing the defect area and features, the edge position of the defect feature is located using the edge model;

[0064] S6: Using a sequential similarity detection algorithm to detect the edge position of the defect feature in step S5, matching the defect target of the original battery thin film surface image.

[0065] In this embodiment, first, a small perovskite film is fixed on an image acquisition platform, the relative distance between the camera and the sample is kept fixed, and image data is captured by a CCD camera. During the image acquisition process, the placement platform can be rotated to increase the number of data acquisitions at different angles. 100 RGB film images are used, including normal and defective states. Figure 3 For images of defective states, Figure 4is an image in a normal state. Then, the image features of the battery film surface image are obtained. The purpose of extracting image features is to obtain surface information, including color features and HSV features. Color is the most intuitive feature in the image, which is represented by RGB (red, green, blue) channels. By analyzing the color, the state or potential defects of the film can be preliminarily judged. The HSV model divides color information into hue (H), saturation (S) and brightness (V). The hue value range of each pixel is 0 to 359 degrees, and the saturation and brightness value range of each pixel is 0.0-1.0. It has the following advantages: hue (H) describes the color type (such as red, green, and blue), saturation (S) describes the purity of the color, the higher the purity, the brighter the color, and brightness (V) indicates the brightness of the color. The HSV model is more robust under different lighting conditions and can more accurately reflect color information. The HSV model is obtained by converting the RGB image into a feature graph under the HSV color model.

[0066] The brightness information in the V channel can help distinguish the film from the background. In the images of perovskite solar cell films prepared in the laboratory, there is usually a significant brightness difference between the main part and the background. The V channel, as a direct representation of brightness information, can effectively capture this difference. The HSV color space helps to reduce the interference of color on image processing and can process brightness information alone without considering the impact of color changes on the processing results. Using the V channel can enhance the brightness difference in the image, making the main part easier to distinguish from the background. By setting a threshold, only areas with a certain brightness are retained, thereby extracting the first image.

[0067] In step S3, noise is removed by considering the similarity between pixels in the entire image. The value of each pixel is calculated based on the weighted average of similar neighboring pixels, which retains edge information, avoids edge blurring that may be caused by traditional mean filtering, and enhances image details and contrast. The image after non-local mean filtering is logarithmically transformed. Logarithmic transformation is a nonlinear grayscale transformation. It expands the low grayscale level of the image and compresses the high grayscale level by applying a logarithmic function. The logarithmic transformation can expand the grayscale level of the dark part of the image, making the details of the low grayscale level more obvious. The effect is better when there are more details in the dark part of the image but the contrast is low. Combined with the use of local mean filtering, the logarithmic transformation can compress the dynamic range of high grayscale levels, so that the contrast between different brightness of high grayscale levels is reduced, thereby avoiding the loss of details due to excessive brightness. By enhancing the low grayscale details and compressing the high grayscale dynamic range, the logarithmic transformation can improve the overall contrast of the image, making the visual effect of the image clearer.

[0068] The purpose of connectivity analysis is to convert the target from the pixel level to the connected component level. By detecting the connectivity of adjacent pixels in the image, continuous regions can be identified. By analyzing the characteristics of connected regions (such as area, shape, etc.), potential defective areas can be screened out.

[0069] Among them, step S5 is also called sub-pixel fitting positioning. The main purpose of sub-pixel fitting positioning is to accurately identify the position of the defect feature. First, based on the feature part screened out in the previous step, it is necessary to find the point with the most significant color change in the edge image. This is the location of the edge. The edge model uses a Gaussian function to represent the intensity of a pixel in the image. When performing sub-pixel positioning, we need to convert the points in the local coordinate system to the image coordinate system, and finally solve the model parameters by analyzing the residual. The residual refers to the difference between the actual image intensity and the model predicted intensity. By minimizing this residual, the optimal model parameters can be obtained, thereby accurately locating the position of the defect feature edge.

[0070] The SSDA algorithm can quickly scan the original image and match the template sub-images one by one, which makes the detection process relatively fast and suitable for processing large-scale image data. By comparing the similarity between each pixel in the image and the template, SSDA can effectively identify subtle defects, thereby improving the accuracy of defect detection. The detection results are as follows: Figure 5 and Figure 6 shown.

[0071] Using RGB and HSV models to extract features from images can fully capture color information. By separating hue, saturation and brightness, the robustness of color information under different lighting conditions is enhanced, which helps to more accurately identify the state and potential defects of battery thin films. Non-local mean filtering can effectively remove noise from images while retaining edge information to avoid detail blurring that may be caused by traditional mean filtering. Combined with logarithmic transformation, the overall contrast of battery thin film images is improved, making the dark details of battery thin films more obvious. Through connectivity analysis, the target is converted from pixel level to connected component level, which helps to identify and screen out continuous defective areas and further improve the accuracy of defect detection. The edge model can accurately identify the location of defect features, thereby improving positioning accuracy.

[0072] Preferably, performing connectivity analysis on the second image to screen out defect areas and extract defect features includes:

[0073] According to the specific starting node and ending node of the enhanced image, whether the two points are connected is analyzed, so as to find a single part and extract the foreground part;

[0074] Through the feature information, the threshold is set to filter out the defect area and extract the defect features.

[0075] When performing connectivity analysis on the second image, first select specific starting and ending nodes, which are located at the edge or inside of the defect area, and then use the traversal algorithm to detect whether there is a connected path between the two nodes to determine the connectivity of the defect area. Once the connectivity is confirmed, the area can be marked as the foreground part, and then the corresponding threshold is set by analyzing the characteristic information of the defect area, such as area and grayscale value, to filter out qualified defect areas and exclude noise or irrelevant parts. Finally, based on the screened defect area, the corresponding feature data is extracted, including the shape, size, location and grayscale value of the defect, to provide a basis for subsequent analysis and processing. Through this series of steps, connectivity analysis can effectively identify and extract defects in the image and improve the accuracy and reliability of detection.

[0076] Preferably, the edge position of the defect feature in step S5 is detected using a sequential similarity detection algorithm, and the defect target matching the surface image of the original battery film includes:

[0077] Make a matching template according to the edge position of the defect feature;

[0078] A sequential similarity detection algorithm is used to scan the template sub-images corresponding to the pixels in the original image one by one to match all defect targets.

[0079] In step S6, a sequential similarity detection algorithm is used to detect defects on the surface of the perovskite film. First, multiple matching templates are made according to the extracted different defect features. Each template represents a specific defect style. Then, a sequential similarity detection algorithm (SSDA) is used for matching. The algorithm scans each pixel in the original image one by one and compares it with the corresponding template sub-image. For each position in the image, SSDA compares the pixel value near the position with the pixel value in the template to calculate the similarity. In this way, the algorithm can quickly identify and match all defect targets. The matching process is efficient. The design of the template enables the algorithm to quickly determine whether there are defects and mark the locations of these defects in the image. Through template matching, various defects on the surface of the perovskite film can be accurately identified to ensure the comprehensiveness and accuracy of the detection.

[0080] Preferably, obtaining the battery film surface image includes:

[0081] Acquire the RGB three-channel color image of the battery film surface image, and perform normalization processing to obtain the color characteristics of the battery film surface image;

[0082] The normalization process satisfies the relationship: R′=R / 255, G′=G / 255, B′=B / 255;

[0083] Among them, R represents the red channel, G represents the red channel, B represents the red channel, and k′, G′, and B′ refer to the values ​​after normalization of R, G, and B, respectively.

[0084] Furthermore, the HSV features of the battery film surface image are obtained including:

[0085] Calculate the values ​​of H, S and V to satisfy the relationship:

[0086] V = max(R′, G′, B′)

[0087]

[0088]

[0089] If H is less than 0, add 360 to H;

[0090] Convert H, S and V to between 0 and 255 to obtain HSV features, and satisfy the relationship:

[0091] The HSV features of the battery film surface image are first obtained by calculating the values ​​of H, S and V, where V is the maximum value of each channel after RGB normalization, S is calculated based on the difference between V and the minimum value of the RGB channel, and H is calculated based on the relationship between different V values ​​and RGB channels. During the calculation process, if H is less than 0, 360 degrees are added to ensure the effective range of H, and finally H, S and V are converted to between 0-255 to obtain standardized HSV features. The HSV model clearly divides color information into hue, saturation and brightness, making the expression of color more intuitive; secondly, the HSV feature has good robustness to changes in illumination and can maintain detection consistency under different illumination conditions. The standardized HSV feature improves the effect of subsequent processing steps, facilitates color matching and comparison between different images, and also reduces computational complexity.

[0092] Preferably, the non-local mean filtering satisfies the relationship:

[0093]

[0094] Among them, I(i, j) represents the pixel value of the surface image of the original battery film, Ω represents the search range of the image, and w(x, y, i, j) represents the weight calculated according to the similarity of the image blocks, and satisfies the relationship:

[0095]

[0096] Among them, I(Ω(x, y)) and I(Ω(i, j)) represent the image blocks centered at (x, y) and (i, j), respectively, and h represents the parameter controlling the weight attenuation.

[0097] By considering the contribution of all similar blocks in the image, non-local mean filtering can effectively suppress noise while retaining image details, especially in images with complex textures and structures. Unlike traditional filtering methods that only rely on local neighborhoods, non-local mean filtering integrates similar information far away from the target pixel, so that the context of the entire image can be utilized, thereby improving the denoising effect. The weight calculation is based on the similarity of image blocks, can adapt to the noise characteristics of different regions, and has a strong adaptability to changes in image content.

[0098] Furthermore, the logarithmic transformation satisfies the relationship:

[0099]

[0100] in, represents the image pixel value after non-local mean filtering, represents the pixel value of the image after logarithmic transformation, and c represents a constant value.

[0101] Preferably, the edge model satisfies the relationship:

[0102]

[0103] Among them, M(x, y; A, B, σ, u, v) represents the intensity at the corresponding pixel position, (x, y) represents a pixel position in the image, A represents the intensity of the image background, B represents the peak intensity in the bright area, (u, v) represents the position of the peak in the bright area, and σ 2 represents the Gaussian model variance.

[0104] Assuming that the coordinates of the target feature are (u0, v0), when performing sub-pixel edge positioning, the position to be positioned is in the edge image where the color changes the most from dark to bright, that is, the position corresponding to the maximum change in the derivative, such as Figure 7 As shown. Combined Figure 7 , we can get the position transformation relationship between the edge point in the local coordinate system and the image coordinate system: The expression M(x, y; A, B, σ, u, v) on the left can be expressed as M(x, θ), where x = (x, y) T represents the pixel position, where θ = (A, B, σ, u, v) T is the model parameter of the point. When the model parameter is known, the image feature has been located, that is, the center position of the Gaussian model (u, v). The coordinate position is floating point, that is, sub-pixel level. Finally, the residual is analyzed to solve the parameter: E(θ) = ∑ x∈W (M(x,θ)-I(x)) 2 , Figure 8 is the edge position of the defect feature.

[0105] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0106] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A method for detecting surface defects of battery thin films based on deep learning, characterized in that: The detection method comprises the following steps: S1: Acquire image features of the battery film surface image, including color features and HSV features; S2: extracting the battery film surface image according to the light and dark difference of the V channel image in the HSV feature, removing irrelevant background, and obtaining the first image; S3: performing non-local mean filtering and logarithmic transformation on the first image in sequence to obtain a second image; S4: performing connectivity analysis on the second image to screen out defect areas and extract defect features; S5: By analyzing the defect area and features, the edge position of the defect feature is located using the edge model; S6: Using a sequential similarity detection algorithm to detect the edge position of the defect feature in step S5, matching the defect target of the original battery thin film surface image.

2. The detection method according to claim 1, characterized in that: Connectivity analysis is performed on the second image to screen out defect areas and extract defect features including: According to the specific starting node and ending node of the enhanced image, whether the two points are connected is analyzed, so as to find a single part and extract the foreground part; Through the feature information, the threshold is set to filter out the defect area and extract the defect features.

3. The detection method according to claim 1, characterized in that: The edge position of the defect feature in step S5 is detected by using a sequential similarity detection algorithm, and the defect targets matching the surface image of the original battery film include: Make a matching template according to the edge position of the defect feature; A sequential similarity detection algorithm is used to scan the template sub-images corresponding to the pixels in the original image one by one to match all defect targets.

4. The detection method according to claim 1, characterized in that: The following are the methods for obtaining images of the battery film surface: Acquire the RGB three-channel color image of the battery film surface image, and perform normalization processing to obtain the color characteristics of the battery film surface image; The normalization process satisfies the relationship: R′=R / 255, G′=G / 255, B′=B / 255; Among them, R represents the red channel, G represents the red channel, B represents the red channel, and R′, G′, and B′ refer to the normalized values ​​of R, G, and B, respectively.

5. The detection method according to claim 4, characterized in that: The HSV features for obtaining images of the battery film surface include: Calculate the values ​​of H, S and V to satisfy the relationship: V = max(R′, G′, B′) If H is less than 0, add 360 to H; Convert H, S and V to between 0 and 255 to obtain HSV features, and satisfy the relationship:

6. The detection method according to claim 1, characterized in that: The non-local mean filtering satisfies the relationship: Among them, I(i, j) represents the pixel value of the surface image of the original battery film, Ω represents the search range of the image, and w(x, y, i, j) represents the weight calculated according to the similarity of the image blocks, and satisfies the relationship: Among them, I(Ω(x, y)) and I(Ω(i, j)) represent the image blocks centered at (x, y) and (i, j), respectively, and h represents the parameter controlling the weight attenuation.

7. The detection method according to claim 6, characterized in that: The logarithmic transformation satisfies the relationship: in, represents the image pixel value after non-local mean filtering, represents the pixel value of the image after logarithmic transformation, and c represents a constant value.

8. The detection method according to claim 1, characterized in that: The edge model satisfies the relationship: Among them, M(x, y; A, B, σ, u, v) represents the intensity at the corresponding pixel position, (x, y) represents a pixel position in the image, A represents the intensity of the image background, B represents the peak intensity in the bright area, (u, v) represents the position of the peak in the bright area, and σ 2 represents the Gaussian model variance.

Images