Lithium battery thermal runaway single image defogging method based on fusion method

Abstract

The invention discloses a lithium battery thermal runaway single image defogging method based on a fusion method, and relates to the technical field of lithium battery image processing, and the method comprises the steps: obtaining a single original image containing smoke interference generated in a lithium battery thermal runaway process; performing adaptive exposure processing on the original image to generate a plurality of underexposure images with different exposure degrees; self-adaptive histogram equalization processing with limited contrast is carried out on the underexposure image; generating a weight map based on each processed image; a Laplacian pyramid fusion method is used; and the edge and the detail area of the image are accurately optimized by using an adaptive detail enhancement algorithm, so that the defogged image is ensured to maintain high-quality detail expression while noise suppression is performed. The finally output image not only has high contrast and definition, but also effectively retains key information in an original scene, and provides more reliable and efficient visual data support for state monitoring of the lithium battery.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery image processing, and in particular to a method for defogging a single image of a lithium battery with thermal runaway based on a fusion method. Background Art

[0002] With the widespread use of lithium batteries, their safety issues have become a focus of attention. Lithium batteries may experience thermal runaway under extreme conditions such as high temperature, overcharging, and collision, causing the internal temperature of the battery to rise sharply, and then cause combustion or explosion. This not only poses a serious threat to the equipment, but also poses a great hidden danger to personal safety. Therefore, it is particularly important to accurately monitor the state changes of lithium batteries during thermal runaway, issue warnings in time, and take effective measures.

[0003] During the thermal runaway of lithium batteries, a large amount of smoke, flames and high-temperature airflow will be generated rapidly. These factors will seriously affect the quality of images captured by monitoring equipment such as cameras, making the images blurred or blocked. Traditional image defogging or enhancement technologies often require multiple images to be processed or rely on prior knowledge of specific scenes, and cannot cope with complex single-image scenes. In practical applications, single-image defogging technology that can clearly capture the thermal runaway process of lithium batteries has important practical value.

[0004] In the existing technology, most of the defogging solutions for single images are limited to defogging in natural environments, such as atmospheric scattering models, dark color priors and other methods. However, these methods have limited performance in complex environments with high temperatures and high smoke concentrations. Because the smoke characteristics generated by thermal runaway are quite different from the haze in the natural environment, the optical properties of different wavelengths and the morphology of smoke particles make ordinary defogging technology ineffective in such applications.

[0005] This patent takes full account of the thermal runaway situation, optimizes and improves the (AMEF) algorithm based on an artificial multi-exposure image fusion method, and proposes a single image defogging method for lithium battery thermal runaway based on the fusion method. This method can effectively improve the defogging effect and obtain more image details and more vivid picture colors without distortion. This patent can be used in lithium battery monitoring systems to provide clear image information, help users understand the status changes of the battery in a timely manner, and thus improve safety.

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0007] The purpose of the present invention is to provide a single image defogging method for lithium battery thermal runaway based on a fusion method. Based on the technical solution, the defogging method effectively removes the image blurring problem caused by interference such as smoke and heat flow during the thermal runaway of lithium batteries by fusing multiple images with different exposures, combining contrast-limited adaptive histogram equalization technology and multi-scale fusion pyramid image processing method. In addition, the edges and detail areas of the image are accurately optimized using an adaptive detail enhancement algorithm to ensure that the defogging image maintains high-quality detail performance while suppressing noise. The final output image not only has high contrast and clarity, but also effectively retains the key information in the original scene, providing more reliable and efficient visual data support for lithium battery status monitoring.

[0008] In order to achieve the above object, the present invention provides the following technical solution: a single image defogging method for lithium battery thermal runaway based on a fusion method, comprising the following steps:

[0009] Acquire a single original image containing smoke interference generated during the thermal runaway process of the lithium battery;

[0010] Performing adaptive exposure processing on the original image to generate a plurality of underexposed images with different exposure levels;

[0011] performing contrast-limited adaptive histogram equalization processing on the underexposed image;

[0012] Generate a weight map based on each processed image;

[0013] Using a Laplacian pyramid fusion method, multi-scale fusion is performed on the weight map and the processed image to generate a fused image;

[0014] Performing adaptive detail enhancement processing on the fused image to enhance detail information in the image;

[0015] The final clear image without smoke interference is output for monitoring the thermal runaway process of lithium batteries.

[0016] Preferably, the specific steps of obtaining a single original image containing smoke interference generated during the thermal runaway process of the lithium battery are as follows: during the thermal runaway process of the lithium battery, due to interference factors such as smoke, flame, and heat flow, it is difficult for conventional image processing to accurately capture key states; this step uses a surveillance camera to capture a single image containing smoke interference at a certain moment during the thermal runaway process: hazy (x,y)=Capture(t 0 ,x,y), where x, y are the spatial coordinates of the image; t 0 is the image capture moment; hazy (x, y) is the captured disturbed image.

[0017] Preferably, the specific steps of performing adaptive exposure processing on the original image to generate a plurality of underexposed images with different exposure degrees are: generating a plurality of images with different exposure degrees by using different exposure coefficients. i (x, y), thereby enhancing the details in the image and increasing the brightness of the image; first set the initial exposure factor α 1 =1, keep the original brightness of the image; then gradually reduce the exposure factor α in a decreasing manner i , generate images with different exposures: α i+1 =α i ×γ, γ<1, i=1,2,3..., where γ is set to 0.8 to ensure that multiple images with different exposures are generated; then perform exposure adjustment: I i (x,y)=α i I hazy (x,y), where α i is the exposure factor, I hazy (x,y) is the original input image.

[0018] Preferably, the specific steps of performing contrast-limited adaptive histogram equalization processing on the underexposed image are: for each image with different exposure, performing contrast-limited adaptive histogram equalization processing, enhancing the local contrast by limiting the histogram equalization, thereby improving the details of the dark and bright areas; first, each image I i (x, y) is divided into several sub-blocks of size P×P, and histogram equalization is performed separately in each sub-block; set P=8, the sub-block size affects the local enhancement effect; then calculate the grayscale histogram h(k) for each sub-block, and then calculate the cumulative distribution function (CDF) c(k): Among them, k is the gray value of the pixel, and h(k) is the frequency of the gray value in the image. The gray value of each pixel is equalized through CDF:

[0019]

[0020] Among them, c max and c min are the maximum and minimum values ​​of CDF in the local area, L max and L min is the maximum and minimum brightness of the area; limit the histogram to avoid excessive contrast enhancement; set the contrast clipping limit clipLimit and clip the frequencies exceeding the limit; h cl ip (k)=min(h(k), clipLimit), clipLimit is set to 2.0.

[0021] Preferably, a weight map W is generated for each processed image i (x, y), used for subsequent multi-scale fusion, the weight map is calculated based on brightness, saturation and local contrast: Brightness calculation: Calculate the brightness L(x, y) by converting RGB to grayscale values. The formula is: L i (x,y)=0.299R(x,y)+0.587G(x,y)+0.114B(x,y), where R, G, and B are the red, green, and blue channel values, respectively; Saturation calculation: Saturation is calculated by the maximum and minimum value difference of RGB:

[0022]

[0023] Local contrast calculation: Local contrast is calculated using the Laplacian operator to measure edge strength:

[0024]

[0025] Where W is the local window, μ i is the mean brightness within the window; the formula for generating the weight map is: W i (x,y)=L i (x,y)·S i (x,y)·C i (x,y).

[0026] Preferably, the images with different exposures are multi-scale fused by Laplacian pyramid to extract the effective information area in each exposure image. The specific operation steps are as follows: constructing Gaussian pyramid: for image I i ′ (x, y) is Gaussian blurred to generate a multi-layer Gaussian pyramid. The formula is: (x, y) = Gaussian (I i ′(x,y),σ l ), where σ l is the standard deviation of the blur kernel, l is the number of pyramid layers; Construct the Laplace pyramid: Construct the Laplace pyramid through the hierarchical difference of the Gaussian pyramid, the formula is: Multi-scale fusion: using weight map W i (x,y) fuses each layer of the Laplacian pyramid: Reconstruct the image: Finally, reconstruct the fused image layer by layer through upsampling. The formula is:

[0027] Preferably, the fused image R is further improved by a detail enhancement algorithm. sThe edge and detail parts of (x, y) are optimized to optimize the sharpness and clarity of the image. The specific operation steps are as follows: The edge information in the enhanced image is calculated by the Laplace operator:

[0028]

[0029] ,in, is the image R s (x, y); then, in order to enhance the detail information of the image, the result of the Laplacian operator is added to the original image, and an enhancement coefficient λ is introduced. This coefficient is used to control the strength of detail enhancement. The formula is: R enhanced (x,y)=R s (x,y)+λ· Among them, R enhanced (x, y) is the image after detail enhancement; λ is a parameter that controls the strength of detail enhancement, which generally ranges from 0.5 to 1.5, depending on the sharpness requirement of the image; Explanation: By adding the enhancement term after the Laplace operator, the sharpness of the image can be effectively improved, the details and edges can be highlighted, and the overall clarity of the image can be improved; Finally, the adaptively adjusted λ value is used to adjust the enhancement strength according to the characteristics of different areas; For example, a smaller λ can be used in the smooth area of ​​the image, and a larger λ value can be used in the edge area, so as to avoid noise amplification caused by over-enhancement. The formula is expressed as:

[0030]

[0031] Where α is the baseline enhancement strength.

[0032] Preferably, the image R after the adaptive detail enhancement processing is enhanced (x,y) output is the final dehazed image R final (x, y), the specific operation steps are as follows: after multi-exposure fusion and detail enhancement processing, the generated image has the effect of removing smoke interference and richer details; at this time, the detail-enhanced image R enhanced As the final dehazed image output: R final (x,y)=R enhanced (x, y), before output, the image can be slightly smoothed using a Gaussian filter to suppress artifacts and noise amplification caused by over-sharpening. The Gaussian smoothing formula is: R final (x,y)=Gaussian(R enhanced (x, y), σ); where σ is the standard deviation of the Gaussian filter, which is generally between 0.5 and 1.0, and Gaussian (·) is the Gaussian filter operation; the image R finalThe noise interference caused by smoke and blur is removed. After detail enhancement, the bright and dark details of the original image are retained, and the edges are clearer.

[0033] In the above technical solution, the technical effects and advantages provided by the present invention are as follows: In the technical solution of the present invention, a method for image defogging based on the integration of multiple technical means is proposed, especially for the image blurring problem caused by interference factors such as smoke and heat flow during the thermal runaway of lithium batteries. Through different innovative steps, the method enables the final generated image to have higher clarity, contrast and detail retention, significantly improving the image quality in complex environments.

[0034] First, this method uses multi-exposure image processing technology. By adjusting the exposure of the original image to different degrees, multiple images with different exposure levels are generated. The advantage of this technology is that it can extract the bright and dark details of the image from different exposure levels, avoiding the problem that a single exposure method is difficult to take into account both brightness and contrast. In a smoky and complex lighting environment, multi-exposure images can capture key details well, laying a solid foundation for subsequent image processing steps.

[0035] Secondly, based on each image with different exposures, contrast-limited adaptive histogram equalization (CLAHE) processing was performed. Traditional histogram equalization methods are prone to overexposure or loss of dark details, while CLAHE can effectively enhance the local contrast of the image by processing in blocks and limiting the contrast enhancement. Especially in high-noise environments, CLAHE can significantly improve the dark details of the image while avoiding noise amplification caused by excessive contrast enhancement. This step not only improves the visual effect of the image, but also improves the performance of the image in complex environments, ensuring the accurate presentation of key details.

[0036] Furthermore, the multi-scale fusion technology of Laplacian pyramid is used to fuse multiple images with different exposures at multiple levels. By weighted fusion of the details of each layer of images, especially the weight map generated based on the brightness, saturation and contrast of the image, it can ensure that the high-information areas in each image are fully expressed. The multi-scale characteristics of Laplacian pyramid ensure the extraction and retention of details at different resolutions, making the details of the bright, dark and edge areas of the fused image more balanced and the visual effect more natural.

[0037] After the image fusion is completed, this method also applies an adaptive detail enhancement algorithm to further optimize the edge and high-frequency information of the image. By using the Laplacian operator to detect the edge of the image and combining the adaptive enhancement coefficient to perform targeted detail processing on different areas, it ensures that the key edge information in the image is effectively retained and enhanced. In particular, the adaptive enhancement coefficient can be adjusted according to the edge strength of different areas, thereby avoiding over-sharpening and noise amplification in smooth areas. This detail enhancement process makes the image clearer and sharper in the edge area while maintaining the overall visual coherence.

[0038] Finally, the dehazed image generated after all the processing steps not only effectively removes the interference of smoke, but also retains rich scene details. Compared with traditional dehazing methods, the present invention performs better in high-noise and complex lighting environments. Especially in the process of thermal runaway of lithium batteries, this method can provide more accurate visual data and provide clearer images for the monitoring system, making the state monitoring and fault detection of lithium batteries more reliable. In addition, through multi-scale fusion and detail enhancement processing of images, the dehazed image maintains a high clarity and stability at different resolutions, providing solid data support for subsequent analysis and processing. In general, the image dehazing method of the invention has broad application prospects in the field of lithium battery safety monitoring, and can significantly improve the image quality and monitoring effect in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present application and the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0040] Figure 1 The present invention is a flow chart of the method. DETAILED DESCRIPTION

[0041] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0042] The present invention provides Figure 1 The single image defogging method for lithium battery thermal runaway based on the fusion method shown includes the following steps: obtaining a single original image containing smoke interference generated during the thermal runaway of the lithium battery;

[0043] Performing adaptive exposure processing on the original image to generate a plurality of underexposed images with different exposure levels;

[0044] performing contrast-limited adaptive histogram equalization processing on the underexposed image;

[0045] Generate a weight map based on each processed image;

[0046] Using a Laplacian pyramid fusion method, multi-scale fusion is performed on the weight map and the processed image to generate a fused image;

[0047] Performing adaptive detail enhancement processing on the fused image to enhance detail information in the image;

[0048] Output the final clear image without smoke interference, which is used for monitoring the thermal runaway process of lithium batteries. The specific steps of obtaining a single original image containing smoke interference generated during the thermal runaway process of lithium batteries are as follows: During the thermal runaway process of lithium batteries, due to interference factors such as smoke, flame, and heat flow, conventional image processing is difficult to accurately capture the key state; this step uses a surveillance camera to capture a single image containing smoke interference at a certain moment during the thermal runaway process: I hazy (x,y)=Capture(t 0 ,x,y), where x, y are the spatial coordinates of the image; t 0 is the image capture moment; hazy (x, y) is the captured disturbed image; in the implementation process, it is necessary to first obtain a single image containing smoke interference. This image usually comes from the monitoring equipment during the thermal runaway of lithium batteries. The characteristics of this type of image are that due to the interference of factors such as smoke, flames and high temperature, the image quality is often poor, manifested as blur, brightness attenuation, contrast reduction and color distortion. During the thermal runaway of lithium batteries, the smoke particles generated cause the outline of objects in the image to be blurred and the details to be difficult to distinguish through the scattering and occlusion effects of light. In such a scene, it is difficult for ordinary image processing methods to restore the key details in the image, and the defogging of a single image requires specially designed processing steps for such highly disturbed images. This process first requires capturing such low-quality single images from the monitoring system as input for subsequent processing. The captured image usually contains a lot of smoke interference, which is caused by the scattering and absorption effects of smoke particles on light. Therefore, after acquiring the image, the system first conducts a preliminary analysis of the overall brightness, contrast and color characteristics of the image to evaluate the severity of the image interference and provide a basis for subsequent defogging.

[0049] The specific steps of performing adaptive exposure processing on the original image to generate multiple under-exposed images with different exposure degrees are as follows: generating multiple images with different exposure degrees by using different exposure coefficients. i (x, y), thereby enhancing the details in the image and increasing the brightness of the image; first set the initial exposure factor α 1 =1, keep the original brightness of the image; then gradually reduce the exposure factor α in a decreasing manner i , generate images with different exposures: α i+1 =α i ×γ, γ<1, i=1,2,3..., where γ is set to 0.8 to ensure that multiple images with different exposures are generated; then perform exposure adjustment: I i (x,y)=α i I hazy (x,y), where α i is the exposure factor, I hazy (x,y) is the input original image; after the image is captured, the next step is to perform adaptive exposure processing. The main purpose of this processing is to perform adaptive exposure adjustment for different brightness areas, so as to optimize the bright and dark details in the image. Generally speaking, the interference of smoke on the image will lead to the non-uniformity of the overall brightness, and some areas may appear too bright, while other areas may appear dim. Adaptive exposure technology can effectively solve this problem by adjusting the brightness range of the image so that the bright areas of the image are not overexposed and the dark areas are not completely obscured. In actual operation, the exposure adjustment of the image is adaptively performed according to the brightness information of different areas. For brighter areas, the system will reduce the exposure intensity to ensure that these areas will not be overexposed or distorted; for darker areas, the system will increase the exposure to reveal hidden details. Through such processing, images with different exposures can retain the key details of the bright and dark areas respectively, thus laying the foundation for subsequent dehazing processing. The advantage of adaptive exposure processing is that it can dynamically adjust to complex scenes without relying on the input of multiple images. It can optimize the brightness and contrast in the image simply by adaptively processing different areas of a single image.

[0050] The specific steps of performing contrast-limited adaptive histogram equalization processing on the underexposed image are as follows: for each image with different exposure, performing contrast-limited adaptive histogram equalization processing, enhancing the local contrast by limiting the histogram equalization, thereby improving the details of the dark and bright areas; first, each image I i(x, y) is divided into several sub-blocks of size P×P, and histogram equalization is performed separately in each sub-block; set P=8, the sub-block size affects the local enhancement effect; then calculate the grayscale histogram h(k) for each sub-block, and then calculate the cumulative distribution function (CDF) c(k): Among them, k is the gray value of the pixel, and h(k) is the frequency of the gray value in the image. The gray value of each pixel is equalized through CDF:

[0051]

[0052] Among them, c max and c min are the maximum and minimum values ​​of CDF in the local area, L max and L min is the maximum and minimum brightness of the area; limit the histogram to avoid excessive contrast enhancement; set the contrast clipping limit clipLimit and clip the frequencies exceeding the limit; h cl ip (k) = min(h(k), clipLimit), clipLimit is set to 2.0; after the adaptive exposure process is completed, CLAHE processing is applied to the image, that is, contrast-limited adaptive histogram equalization. CLAHE technology is an image processing algorithm used to enhance the local contrast of an image, especially for high-noise, low-contrast scenes such as thermal runaway. Traditional histogram equalization can easily lead to over-enhancement of some areas in the image, while CLAHE can prevent overexposure in areas with higher brightness through a contrast limitation mechanism, while improving the detail performance in dark areas. Specifically, CLAHE processing divides the image into several local areas and performs histogram equalization operations separately in each area. This localized processing can better preserve local details in the image and avoid the loss of details that may be caused by global enhancement. In dark areas, CLAHE can enhance the contrast of low-brightness pixels and reveal detail information that was originally difficult to distinguish; while in bright areas, CLAHE limits over-enhancement to ensure that the details in the highlight areas are not distorted due to excessive contrast. Finally, the image processed by CLAHE has a more balanced brightness distribution and clearer local details, which is significantly effective in removing smoke interference and improving image quality.

[0053] Generate a weight map W for each processed image i (x, y), used for subsequent multi-scale fusion, the weight map is calculated based on brightness, saturation and local contrast: Brightness calculation: Calculate the brightness L(x, y) by converting RGB to grayscale values. The formula is: L i(x,y)=0.299R(x,y)+0.587G(x,y)+0.114B(x,y), where R, G, and B are the red, green, and blue channel values, respectively; Saturation calculation: Saturation is calculated by the maximum and minimum value difference of RGB:

[0054]

[0055] Local contrast calculation: Local contrast is calculated using the Laplacian operator to measure edge strength:

[0056] Where W is the local window, μ i is the mean brightness within the window; the formula for generating the weight map is: W i (x,y)=L i (x,y)·S i (x,y)·C i (x,y); after CLAHE processing, a weight map for each image needs to be generated. The role of the weight map is to determine the importance of different image regions in the subsequent image fusion process, so as to highlight the key details in the image. Specifically, the generation of the weight map is based on three key factors: brightness, saturation, and local contrast. Brightness represents the light intensity of each pixel in the image. Usually, higher brightness means important information areas in the image. Therefore, when generating the weight map, higher brightness areas will be given larger weights. Saturation reflects the purity of the color in the image. Higher saturation usually means that the image has richer color information, so these areas will also occupy a larger weight in the weight map. Local contrast is calculated by the Laplacian operator, which can reflect the intensity of the edges and details in the image. In the image fusion process, edge areas and high-contrast areas usually contain a lot of important information, so areas with higher local contrast will also receive larger weights. By combining the weight map generated by brightness, saturation, and local contrast, it can be ensured that in the subsequent image processing steps, the key information areas of the image are fully retained, while areas with more noise and interference are weakened.

[0057] The Laplace pyramid is used to perform multi-scale fusion of images with different exposures, and the effective information area in each exposure image is extracted. The specific operation steps are as follows: Construct a Gaussian pyramid: i ′ (x, y) is Gaussian blurred to generate a multi-layer Gaussian pyramid. The formula is: (x, y) = Gaussian (I i ′(x,y),σ l ), where σ l is the standard deviation of the blur kernel, l is the number of pyramid layers; Construct the Laplace pyramid: Construct the Laplace pyramid through the hierarchical difference of the Gaussian pyramid, the formula is: Multi-scale fusion: using weight map W i (x,y) fuses each layer of the Laplacian pyramid: Reconstruct the image: Finally, reconstruct the fused image layer by layer through upsampling. The formula is: After the weight map is generated, the Laplacian pyramid fusion technology is used for multi-scale image fusion. Laplacian pyramid fusion is a multi-scale processing method that decomposes the image into multiple levels of different resolutions and fuses them layer by layer to retain the detail information in each level. Specifically, the image is first Gaussian blurred to generate a series of image pyramids with different resolutions. Each layer represents the details of the image at different scales. Then, by calculating the difference of each layer of the image, the corresponding Laplacian pyramid is generated. These pyramid levels contain the edge and detail information of the image at different scales. Next, according to the weight map generated previously, the image of each layer is weighted fused. The high-weight area will be more retained, while the low-weight area will be weakened. The advantage of Laplacian pyramid fusion is that it can fuse images at multiple resolution levels to ensure that the details of the image at different scales can be effectively extracted and retained. Finally, after multi-scale fusion, the generated image not only has the defogging effect, but also retains the rich details in the image, making the image clearer and more natural in visual effect.

[0058] The fused image R is further improved by the detail enhancement algorithm. s The edge and detail parts of (x, y) are optimized to optimize the sharpness and clarity of the image. The specific operation steps are as follows: The edge information in the enhanced image is calculated by the Laplace operator:

[0059]

[0060] ,in, is the image R s (x, y); then, in order to enhance the detail information of the image, the result of the Laplacian operator is added to the original image, and an enhancement coefficient λ is introduced. This coefficient is used to control the strength of detail enhancement. The formula is: R enhanced (x,y)=R s (x,y)+λ· Among them, R enhanced(x, y) is the image after detail enhancement; λ is a parameter that controls the strength of detail enhancement, which generally ranges from 0.5 to 1.5, depending on the sharpness requirement of the image; Explanation: By adding the enhancement term after the Laplace operator, the sharpness of the image can be effectively improved, the details and edges can be highlighted, and the overall clarity of the image can be improved; Finally, the adaptively adjusted λ value is used to adjust the enhancement strength according to the characteristics of different areas; For example, a smaller λ can be used in the smooth area of ​​the image, and a larger λ value can be used in the edge area, so as to avoid noise amplification caused by over-enhancement. The formula is expressed as:

[0061]

[0062] Among them, α is the baseline enhancement strength; after completing the multi-scale fusion of the Laplacian pyramid, the next step is to perform adaptive detail enhancement processing. The main purpose of this process is to further improve the sharpness and detail performance of the image, especially the clarity of the edge part. Adaptive detail enhancement detects high-frequency information and edge areas in the image and focuses on enhancing these areas to ensure that the key details in the image are effectively preserved. Specifically, the detail enhancement algorithm first detects the edge information in the image through the Laplacian operator. The Laplacian operator is a second-order derivative operator commonly used in image processing, which can highlight the edges and high-frequency parts of the image. After detecting the edge information, the system will adaptively adjust the intensity of detail enhancement according to the intensity of the edge area. For areas with strong edges, the system will increase the enhancement coefficient to further improve the sharpness of these areas; while for smooth areas, the system will reduce the enhancement intensity to avoid noise amplification caused by excessive enhancement. Through this adaptive enhancement method, the edge part of the image is clearer, and the area with large noise is effectively suppressed, ensuring the overall balance of the image.

[0063] The image R after adaptive detail enhancement enhanced (x,y) output is the final dehazed image R final (x, y), the specific operation steps are as follows: after multi-exposure fusion and detail enhancement processing, the generated image has the effect of removing smoke interference and richer details; at this time, the detail-enhanced image R enhanced As the final dehazed image output: R final (x,y)=R enhanced (x, y), before output, the image can be slightly smoothed using a Gaussian filter to suppress artifacts and noise amplification caused by over-sharpening. The Gaussian smoothing formula is: R final (x,y)=Gaussian(R enhanced(x, y), σ); where σ is the standard deviation of the Gaussian filter, which is generally between 0.5 and 1.0, and Gaussian (·) is the Gaussian filter operation; the image R final The noise interference caused by smoke and blur is removed. After detail enhancement, the bright and dark details of the original image are retained, and the edges are clearer; after adaptive detail enhancement, the edge clarity and overall visual effect of the image are significantly improved. The defogging image generated at this time has high contrast, detail clarity and noise suppression capabilities. In order to further optimize the output effect of the image, the image can be slightly smoothed by a Gaussian filter to reduce possible artifacts and noise amplification caused by detail enhancement. The function of the Gaussian filter is to smooth the image, eliminate some overly sharp edges and noise, and make the image look more natural. After this series of processing, the final defogging image not only removes the interference caused by smoke during the thermal runaway of the lithium battery, but also retains the key detail information in the original scene, with high visual clarity and quality. The defogging image can provide more reliable visual data support for the state monitoring of lithium batteries, and due to the multi-scale fusion and adaptive detail enhancement in the image processing process, the image maintains high clarity and stability at different resolutions. This defogging technology is particularly suitable for image processing in complex environments such as thermal runaway of lithium batteries, and can significantly improve the reliability and effectiveness of the monitoring system.

[0064] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A single image defogging method for lithium battery thermal runaway based on a fusion method, characterized by: The following steps are involved: Acquire a single original image containing smoke interference generated during the thermal runaway process of the lithium battery; Performing adaptive exposure processing on the original image to generate a plurality of underexposed images with different exposure levels; performing contrast-limited adaptive histogram equalization processing on the underexposed image; Generate a weight map based on each processed image; Using a Laplacian pyramid fusion method, multi-scale fusion is performed on the weight map and the processed image to generate a fused image; Performing adaptive detail enhancement processing on the fused image to enhance detail information in the image; The final clear image without smoke interference is output for monitoring the thermal runaway process of lithium batteries.

2. According to the method for defogging a single image of a lithium battery thermal runaway based on a fusion method according to claim 1, the specific steps of obtaining a single original image containing smoke interference generated during the thermal runaway process of the lithium battery are as follows: during the thermal runaway process of the lithium battery, due to interference factors such as smoke, flame, and heat flow, conventional image processing is difficult to accurately capture key states; this step uses a surveillance camera to capture a single image containing smoke interference at a certain moment during the thermal runaway process: hazy (x,y)=Capture(t0,x,y), where x, y are the spatial coordinates of the image; t0 is the moment of image capture; I hazy (x,y) is the captured disturbed image.

3. According to the single image defogging method for lithium battery thermal runaway based on the fusion method of claim 1, the specific steps of performing adaptive exposure processing on the original image to generate multiple under-exposed images with different exposure degrees are: generating multiple images with different exposure degrees by using different exposure coefficients. i (x, y), thereby enhancing the details in the image and increasing the brightness of the image; first set the initial exposure coefficient α1 = 1 to maintain the original brightness of the image; then gradually reduce the exposure coefficient α by decreasing i , generate images with different exposures: α i+1 =α i ×γ, γ<1,i=1,2,3..., where, γ is set to 0.8 to ensure that multiple images with different exposures are generated; then adjust the exposure: I i (x,y)=α i I hazy (x,y), where α i is the exposure factor, I hazy (x,y) is the original input image.

4. According to the single image defogging method for lithium battery thermal runaway based on the fusion method of claim 1, the specific steps of performing contrast-limited adaptive histogram equalization processing on the underexposed image are: for each image with different exposure, performing contrast-limited adaptive histogram equalization processing, enhancing the local contrast by limiting the histogram equalization, thereby improving the details of the dark and bright areas; first, each image I i (x, y) is divided into several sub-blocks of size P×P, and histogram equalization is performed separately in each sub-block; set P=8, the sub-block size affects the local enhancement effect; then calculate the grayscale histogram h(k) for each sub-block, and then calculate the cumulative distribution function (CDF) c(k): in, k is the gray value of the pixel, h(k) is the frequency of the gray value in the image; the gray value of each pixel is equalized through CDF: are the maximum and minimum values ​​of the brightness of the area; Limit the histogram to avoid excessive contrast enhancement; set the contrast clipping limit clipLimit and clip the frequencies exceeding the limit; h clip (k)=min(h(k), clipLimit), clipLimit is set to 2.

0.

5. According to the single image defogging method for lithium battery thermal runaway based on the fusion method of claim 1, a weight map W is generated for each processed image. i (x, y), used for subsequent multi-scale fusion, the weight map is calculated based on brightness, saturation and local contrast: Brightness calculation: Calculate the brightness L(x, y) by converting RGB to grayscale values. The formula is: L i (x,y)=0.299R(x,y)+0.587G(x,y)+0.114B(x,y), where: R, G, B are red, green, and blue channel values ​​respectively; Saturation calculation: Saturation is calculated by the maximum and minimum value difference of RGB: Local contrast calculation: Local contrast is calculated using the Laplacian operator to measure edge strength: Where W is the local window, μ i is the mean brightness within the window; the formula for generating the weight map is: W i (x,y)=L i (x,y)·S i (x,y)·C i (x,y).

6. According to the single image defogging method for lithium battery thermal runaway based on the fusion method of claim 1, multi-scale fusion is performed on images of different exposures through the Laplace pyramid to extract the effective information area in each exposure image. The specific operation steps are as follows: construct a Gaussian pyramid: i ′ (x, y) is Gaussian blurred to generate a multi-layer Gaussian pyramid. The formula is: (x, y) = Gaussian (I i ′(x,y),σ l ), where σ l is the standard deviation of the blur kernel, l is the number of pyramid layers; Construct the Laplace pyramid: Construct the Laplace pyramid through the hierarchical difference of the Gaussian pyramid, the formula is: Multi-scale fusion: using weight map W i (x,y) fuses each layer of the Laplacian pyramid: Reconstruct the image: Finally, reconstruct the fused image layer by layer through upsampling. The formula is:

7. According to the single image defogging method for lithium battery thermal runaway based on the fusion method of claim 1, the fused image R is further improved by a detail enhancement algorithm. s The edge and detail parts of (x, y) are optimized to optimize the sharpness and clarity of the image. The specific operation steps are as follows: The edge information in the enhanced image is calculated by the Laplace operator: in, is the image R s (x, y); then, in order to enhance the detail information of the image, the result of the Laplacian operator is added to the original image, and an enhancement coefficient λ is introduced. This coefficient is used to control the strength of detail enhancement. The formula is: Among them, R enhanced (x, y) is the image after detail enhancement; λ is a parameter that controls the strength of detail enhancement, which generally ranges from 0.5 to 1.5, depending on the sharpness requirement of the image; Explanation: By adding the enhancement term after the Laplace operator, the sharpness of the image can be effectively improved, the details and edges can be highlighted, and the overall clarity of the image can be improved; Finally, the adaptively adjusted λ value is used to adjust the enhancement strength according to the characteristics of different areas; For example, a smaller λ can be used in the smooth area of ​​the image, and a larger λ value can be used in the edge area, so as to avoid noise amplification caused by over-enhancement. The formula is expressed as: Where α is the baseline enhancement strength.

8. The method for defogging a single image of a lithium battery thermal runaway based on a fusion method according to claim 1, characterized in that: The image R after adaptive detail enhancement enhanced (x,y) output is the final dehazed image R final (x, y), the specific operation steps are as follows: after multi-exposure fusion and detail enhancement processing, the generated image has the effect of removing smoke interference and richer details; at this time, the detail-enhanced image R enhanced As the final dehazed image output: R final (x,y)=R enhanced (x, y), before output, the image can be slightly smoothed using a Gaussian filter to suppress artifacts and noise amplification caused by over-sharpening. The Gaussian smoothing formula is: R final (x,y)=Gaussian(R enhanced (x, y), σ); where σ is the standard deviation of the Gaussian filter, which is generally between 0.5 and 1.0, and Gaussian (·) is the Gaussian filter operation; the image R final The noise interference caused by smoke and blur is removed. After detail enhancement, the bright and dark details of the original image are retained, and the edges are clearer.

Images