A method for processing infrared thermal imaging images for painful areas

Abstract

The present invention proposes a method for processing infrared thermal imaging images for painful areas, which relates to the field of image enhancement. The method combines multi-scale adaptive contrast enhancement and a bidirectional loop mechanism to capture the global and local temperature features of infrared images. An iterative feature enhancement module is constructed through an incremental fusion strategy, and different levels of features are fused in stages. A learnable parameter is introduced to dynamically adjust the fusion ratio of new and old features to improve the adaptability and flexibility of the model. A staged training strategy is adopted to enable the model to learn low-level thermal image information and extract high-level structural and thermal anomaly semantic features, thereby improving the clarity of image details and clinical recognition value. This method enhances the model's adaptability to different types of lesion areas and meets the actual needs of efficient and automated processing of infrared images.

Description

Technical Field

[0001] The present invention belongs to the technical field of image enhancement, and in particular relates to an infrared thermal imaging image processing method for a pain area. Background Art

[0002] Infrared thermal imaging image processing of painful areas is of great value in medical applications such as chronic disease management, postoperative monitoring, and neurological function assessment. However, due to factors such as the resolution of infrared thermal imaging equipment, the complexity of skin surface temperature distribution, and environmental interference, thermal images often have problems such as low contrast, blurred edges, and many thermal noise points, which affect the accurate identification and diagnosis of painful areas. Existing image enhancement methods mostly focus on global brightness contrast adjustment, ignoring the need to enhance detailed features such as temperature distribution patterns and regional thermal anomalies, and usually require manual intervention, making it difficult to adapt to automated clinical processes.

[0003] In recent years, deep learning has made significant progress in medical image enhancement, but most existing methods are general designs and lack structural modeling and dynamic enhancement mechanisms for "regional abnormal thermal features" in infrared thermal images. During the enhancement process, it is difficult to balance the feature expression of global and local thermal distribution, which can easily cause blurred boundaries of hot spot areas or distorted details.

[0004] To address these problems, the present invention proposes an infrared thermal imaging image processing method for painful areas, aiming to improve the automation and accuracy of image processing and address the shortcomings of existing methods. Summary of the Invention

[0005] The present invention proposes a processing method for infrared thermal imaging images of pain areas. Through structural vector modeling, region-guided enhancement, multidimensional evaluation and weight coupling mechanism, the adaptability, detail retention ability and regional targeting of image enhancement are improved. This method is suitable for visual enhancement of temperature abnormality areas in thermal images, and is particularly suitable for high-precision identification and enhancement of pain-related temperature abnormality areas in medical environments.

[0006] The present invention aims to propose an infrared thermal imaging image enhancement model and provide a method for processing infrared thermal imaging images of painful areas, which includes the following steps.

[0007] S1. Construct an initial structural description, collect infrared thermal imaging images of the painful area, construct a preliminary abnormal area distribution map based on edge segmentation, and produce a structural vector map.

[0008] S2. According to the structural vector graph, the number, location, size and edge blur degree information of the abnormal areas are extracted to form a structural guidance graph.

[0009] S3. Cutting the infrared thermal imaging image into multiple image blocks, and performing enhancement processing on each image block according to its local features and the structure guidance map.

[0010] S4. Calculate the three indicators of structural similarity, edge continuity and local contrast for the enhanced image block, calculate the quality score, and adjust the enhancement parameters of the next image block according to the score.

[0011] S5. Use the structural frequency-curvature coupling to map the thermal imaging image structure, calculate the image local frequency response, regional curvature value, information entropy, and construct a nonlinear coupling mapping function.

[0012] S6. Merge the image blocks to obtain an overall enhanced thermal imaging image output.

[0013] Preferably, in step S1, an initial structural description is constructed, and infrared thermal imaging images of normal and pathological pain areas are collected, wherein the pathological areas include inflammatory pain, neuropathic pain and postoperative traumatic pain. The regions of interest are annotated for each thermal imaging image, and the regions of interest include abnormally heated edges, core hot spots and peripheral transition zones. A preliminary abnormal thermal distribution map is constructed based on edge segmentation, and a structural vector map of the pain area is produced.

[0014] Preferably, in step S2, the method for constructing the structure guidance graph comprises the following steps:

[0015] The spatial location information, area scale information, and boundary temperature gradient fuzziness index of each pain area are extracted from the abnormal thermal distribution map. The multidimensional thermal features are encoded into a feature combination for structural modeling. The description unit of each pain area contains coordinate information indicating its positional relationship in the infrared thermal imaging image, a scale parameter reflecting the range of its abnormal thermal diffusion, and a fuzziness parameter used to measure the clarity of the edge temperature transition. The set of structural vectors is used to construct the overall structural guidance map. , the specific formula is:

[0016] ;

[0017] Where, It is a structural guide map. It means the first The center coordinates of the pain area ,radius and edge blur , is the number of regions.

[0018] Preferably, in step S2, information such as the center coordinates, area scale, and edge temperature blur of each painful area in the image is vectorized and encoded and organized into a graph structure representation that can be used for regional positioning and structural guidance, namely a structure-guided graph. This graph is used to provide spatial constraints on thermal anomaly features during the enhancement process, enabling each image block to be associated with the corresponding abnormal area, thereby achieving regional adaptability of the enhancement operation. By constructing the structure-guided graph, not only is the key temperature distribution features in the image retained, but the spatial selectivity and directional accuracy of the enhancement process are also improved. This is particularly suitable for image enhancement tasks in areas with multiple thermal distribution morphological differences.

[0019] Preferably, in step S3, the colony image block is cut as follows:

[0020] S31, based on the preset two-dimensional grid division strategy, the input infrared thermal imaging image is evenly divided in the spatial dimension to form a number of non-overlapping sub-image areas, each of which has the same size in the row and column directions, and the division result covers the entire infrared image. Each image block is set with a clear position index mark for regional level matching and correspondence with the structure guide map, so that the infrared image is divided into multiple sub-image areas according to the preset grid size. Cut into multiple equal image blocks, each image block size is ,in and Represent row and column indices respectively;

[0021] S32. The enhancement processing operation adopts an image block enhancement strategy based on feature response guidance. The strategy dynamically adjusts the contrast enhancement amplitude and thermal texture detail retention strength of the image block by analyzing the temperature distribution characteristics, structural guidance information and spatial position relationship of the current infrared image block in the overall infrared thermal imaging image. The enhancement parameters are determined based on the image statistical characteristics of the current block and the characteristics of the corresponding abnormal thermal area in the structural guidance map. Differential enhancement operations are applied to each infrared image block in a block-by-block processing manner. The calculation formula of the enhancement function is:

[0022] ;

[0023] Where, The original infrared thermal imaging image image blocks, It is a structural guide diagram. is the enhancement parameter, is the enhancement function.

[0024] Preferably, in step S3, the entire image is first divided into equal sizes based on a two-dimensional regular grid strategy to generate a number of image sub-blocks with spatial index attributes, ensuring that each image unit can be independently located and enhanced in the subsequent processing process. Then, combined with the regional feature information provided in the structure guidance map, a customized enhancement operation is performed on each image block. During the enhancement process, the structure matching relationship is used to dynamically adjust the brightness, contrast or edge strength processing parameters of the image block. This processing strategy can allocate differentiated enhancement amplitudes in different image areas to achieve the effect of highlighting the detail areas and moderately suppressing the non-target areas, thereby improving the structural clarity and information density of the overall image.

[0025] Preferably, in step S4, the quality score is calculated as follows:

[0026] S41, calculate the structural similarity SSIM of each image block after enhancement, by taking the grayscale mean of the image blocks before and after enhancement and , grayscale variance and the covariance between the two Statistical extraction is performed to construct a comprehensive evaluation model based on brightness consistency, contrast preservation, and structural reconstruction. After normalization and suppression adjustment, an evaluation value representing the structural consistency of the image block is generated. The calculation formula is:

[0027] ;

[0028] Where, and For image blocks Instead of enhancing the average value of image patches, is the covariance of the image block and the enhanced image block, and is a constant;

[0029] S42. The edge continuity index is constructed by analyzing the difference in edge gradients before and after image block enhancement. First, the gradient amplitude map is extracted from each image block, and the gradient response value of each pixel position in the original image and the enhanced image is recorded respectively. Then, the point-by-point difference of the response value before and after the same position is calculated, and the average is taken for the entire image area to obtain the edge disturbance statistics. The edge continuity after enhancement is calculated. The calculation formula is:

[0030] ;

[0031] Where, is the edge continuity score of the image block, For image blocks No. The gradient of pixels, The enhanced image block No. The gradient of pixels, is the number of pixels in the image block to be calculated;

[0032] S43. Calculate the brightness deviation between each pixel in the image block and its neighborhood, and then normalize it to a standard contrast response. This response is determined by the pixel value and the grayscale mean and standard deviation of the image block in which it is located. After performing the above processing on the image block before and after enhancement, calculate the difference between the two, and perform statistical aggregation over the entire block to calculate the local contrast improvement score. The calculation formula is:

[0033] ;

[0034] Where, yes The local contrast improvement score of the image patch, For image blocks No. The gray value of a pixel, The enhanced image block No. The gray value of a pixel, is an image block The mean value of all pixels in , is the enhanced image block The mean value of all pixels in ;

[0035] S44. The fusion score takes the three indicators of structural similarity, edge continuity and local contrast as input, generates the overall image block quality evaluation value based on the multi-factor weighting mechanism, and feeds it back to the parameter adjustment module. Each sub-indicator is assigned a fixed weight coefficient according to its importance in the task. The linear superposition strategy is adopted in the fusion process, and the scale uniformity between the indicators is considered. The calculation formula is:

[0036] ;

[0037] Where, is the weight coefficient.

[0038] Preferably, in step S4, for each enhanced image block, three independent quality evaluation indicators, namely structural similarity, edge continuity and local contrast, are introduced to evaluate whether the enhanced image maintains structural consistency, whether the edge morphology is destroyed, and whether the contrast of the detail area is improved. By calculating the statistical properties of the image block, a quantitative scoring mechanism is established, and the three sub-scores are fused according to the task weights to form a total quality score. The fusion score can not only be used to evaluate the current enhancement effect, but also as a feedback signal to dynamically adjust the enhancement parameters of subsequent image blocks, thereby realizing adaptive adjustment and stable optimization of the enhancement strategy, and improving the enhancement consistency and local detail quality.

[0039] Preferably, in step S5, the calculation method of the infrared thermal imaging image structure frequency curvature coupling map is:

[0040] S51. Perform wavelet transform on the original infrared thermal imaging image to extract its sub-band responses in the horizontal, vertical and diagonal directions, which respectively reflect the texture intensity changes of the infrared thermal imaging image in different directions. For each pixel position, count its response amplitude in the three directional sub-bands, obtain its overall frequency response intensity through the amplitude combination strategy, and calculate the local frequency response of the image. , the formula for calculating the high frequency component of each pixel through wavelet transform is:

[0041] ;

[0042] Where, are the wavelet coefficients of infrared thermal imaging images in horizontal, vertical and diagonal directions respectively;

[0043] S52. By obtaining the gradient values ​​of the infrared thermal imaging image in the horizontal and vertical directions, further calculate its second-order derivative, including the lateral change rate, longitudinal change rate and mixed change rate. In the processing process, a small regularization factor is introduced to calculate the local curvature value. , the calculation formula based on the second-order derivative is:

[0044] ;

[0045] Where, It is The local curvature value at the pixel, is the gradient of the infrared thermal imaging image in the x direction, is the gradient of the infrared thermal imaging image in the y direction, is the second-order derivative of the infrared thermal imaging image in the x direction, is the mixed second-order derivative of the infrared thermal imaging image in the xy direction, is the second-order derivative of the infrared thermal imaging image in the y direction, is a small constant;

[0046] S53, establish a sliding window with each pixel as the center, count the gray value distribution of all pixels in the window, and calculate its probability distribution, construct an entropy function based on the probability density, and add a minimal bias to calculate the information entropy in the process of probability density construction , the calculation formula based on local grayscale distribution is:

[0047] ;

[0048] Where, Grayscale value The probability distribution of is a constant;

[0049] S54. The generation of enhancement weights is based on the three characteristic parameters of structural frequency, local curvature and information entropy, which are integrated through a nonlinear coupling mechanism. First, the scale of the above three types of features is normalized. Then, control factors are allocated according to task requirements to adjust the influence of each parameter on the enhancement weight. Then, control factors are allocated according to task requirements to adjust the influence of each parameter on the enhancement weight. The enhancement parameters are generated through nonlinear coupling mapping. The calculation formula is:

[0050] ;

[0051] Where, is the final enhancement weight.

[0052] Preferably, in step S5, pixel-level features of the image are collected from three complementary dimensions, including frequency response reflecting the activity of high-frequency details, curvature change characterizing edge bending or corner characteristics, and information entropy measuring regional complexity. Each feature calculates the statistical response of the image through a preset window, and after scale unification and feature normalization, it is input into a nonlinear mapping function to generate a pixel-level enhancement response weight. This weight is used to control the processing intensity of each pixel during the enhancement process, to achieve enhanced response in texture-rich areas and suppress interference in background smooth areas, thereby improving the overall image structure recognition rate, detail recovery ability and visual balance, and meeting the requirements of high-precision image enhancement tasks.

[0053] Compared with the prior art, the present invention has the following technical effects:

[0054] The present invention provides a processing method for infrared thermal imaging images of pain areas. By constructing a structural vector map of the pain area, it realizes the quantitative expression of the position, scale and edge blur of the abnormal hot area in the image, and provides precise regional guidance for subsequent image block enhancement. Unlike the traditional method based on a unified enhancement strategy, the present invention supports local enhancement according to the difference of image blocks, and the enhancement results are more delicate and the boundaries are more complete. It introduces a multi-dimensional quality scoring mechanism, and dynamically evaluates each image block based on the comprehensive structural similarity, edge continuity and local contrast. The results are fed back for subsequent enhancement parameter adjustment to construct a closed-loop adaptive enhancement process. Furthermore, the frequency response, curvature information and information entropy are integrated to construct a coupling mapping, generate pixel-level enhancement weights, and achieve a dynamic balance between detail area enhancement and background area suppression, thereby improving the structural clarity and visual consistency of the enhanced image. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a flow chart of infrared thermal imaging image data processing provided by the present invention.

[0056] Figure 2This is a cutting enhancement module diagram of the infrared thermal imaging image block provided by the present invention.

[0057] Figure 3 This is a flow chart of infrared thermal imaging image block enhancement parameters provided by the present invention.

[0058] Figure 4 This is a diagram of a nonlinear coupling mapping function module for infrared thermal imaging images provided by the present invention.

[0059] Figure 5 This is the effect diagram of the infrared thermal imaging image before enhancement provided by the present invention.

[0060] Figure 6 This is an enhanced image effect diagram obtained by processing the infrared thermal imaging image enhancement model provided by the present invention. DETAILED DESCRIPTION

[0061] The present invention aims to propose a processing method for infrared thermal imaging images of painful areas, proposes an infrared thermal imaging image enhancement model, constructs an infrared thermal image dataset, collects infrared images of normal and pathological painful areas, and annotates the regions of interest. The annotation content includes the center, area scale and edge temperature information of the abnormal hot area, and produces an annotated dataset for model training. The local contrast of the image is enhanced through a multi-scale mechanism. The image registration operation is based on transformation invariant features to ensure the alignment of key thermal abnormality areas between images, reduce geometric differences, and improve the stability of the model under different shooting angles or thermal sensitive resolution conditions. Combining local frequency response, curvature value and information entropy features, the enhancement parameters are dynamically adjusted through structural frequency curvature coupling mapping. The local frequency response captures high-frequency texture details in the thermal image through wavelet transform, the curvature value detects edge structure changes through second-order derivatives, and the information entropy is used to evaluate the temperature complexity of the local area. Finally, the enhancement weight is generated through nonlinear coupling mapping to achieve fine control of image details, retain structural features, and significantly improve the detail expression ability and visual clarity of the infrared image.

[0062] See Figure 1 As shown, a method for processing infrared thermal imaging images of painful areas in an embodiment of the present application.

[0063] S1. Construct an initial structural description, collect infrared thermal imaging images of the painful area, construct a preliminary abnormal area distribution map based on edge segmentation, and produce a structural vector map.

[0064] Furthermore, in step S1, infrared thermal imaging images of normal and pathological pain areas are collected. The pathological images include inflammatory pain, neuropathic pain and postoperative traumatic pain areas. The regions of interest are annotated for each thermal imaging image using professional medical image annotation tools. The annotation content includes abnormal temperature edges, core hot spots and transition temperature zones to ensure the precise positioning of key areas. The annotation results are saved in the XML standard format and correspond one-to-one with the original image files to ensure the high quality and integrity of the infrared image dataset.

[0065] S2. According to the structural vector graph, the number, location, size and edge blur degree information of the abnormal areas are extracted to form a structural guidance graph.

[0066] Furthermore, in step S2, the specific steps of the method for constructing the structure guidance graph are as follows.

[0067] The spatial location information, area scale information, and boundary temperature gradient fuzziness index of each pain area are extracted from the abnormal thermal distribution map. The multidimensional thermal features are encoded into a feature combination for structural modeling. The description unit of each pain area contains coordinate information indicating its positional relationship in the infrared thermal imaging image, a scale parameter reflecting the range of its abnormal thermal diffusion, and a fuzziness parameter used to measure the clarity of the edge temperature transition. The set of structural vectors is used to construct the overall structural guidance map. , the specific formula is:

[0068] ;

[0069] Where, It is a structural guide diagram. It means the first The center coordinates of the pain area ,radius and edge blur , It is the number of regions. During the implementation process, the number of regions should be controlled not to exceed 50.

[0070] S3. Cutting the infrared thermal imaging image into multiple image blocks, and performing enhancement processing on each image block according to its local features and the structure guidance map.

[0071] Furthermore, in step S3, Figure 2 As shown in FIG, the specific steps of cutting the colony image block are as follows.

[0072] S31, based on the preset two-dimensional grid division strategy, the input infrared thermal imaging image is evenly divided in the spatial dimension to form a number of non-overlapping sub-image areas, each of which has the same size in the row and column directions, and the division result covers the entire infrared image. Each image block is set with a clear position index mark for regional level matching and correspondence with the structure guide map, so that the infrared image is divided into multiple sub-image areas according to the preset grid size. Cut into multiple equal image blocks, the size of each colony image block is ,in and Represents row and column indexes respectively. During implementation, the length and width of the control grid are controlled to be no greater than 100, and the initial value is 50×50;

[0073] S32. The enhancement processing operation adopts an image block enhancement strategy based on feature response guidance. The strategy dynamically adjusts the contrast enhancement amplitude and thermal texture detail retention strength of the image block by analyzing the temperature distribution characteristics, structural guidance information and spatial position relationship of the current infrared image block in the overall infrared thermal imaging image. The enhancement parameters are determined based on the image statistical characteristics of the current block and the characteristics of the corresponding abnormal thermal area in the structural guidance map. Differential enhancement operations are applied to each infrared image block in a block-by-block processing manner. The calculation formula of the enhancement function is:

[0074] ;

[0075] Where, The original infrared thermal imaging image image blocks, It is a structural guide diagram. The enhancement parameter range is [0.5, 2.0], is the enhancement function.

[0076] S4. Calculate the three indicators of structural similarity, edge continuity and local contrast for the enhanced image block, calculate the quality score, and adjust the enhancement parameters of the next image block according to the score.

[0077] Further, in step S4, as Figure 3 As shown, the quality score is calculated as follows.

[0078] S41, calculate the structural similarity SSIM of each image block after enhancement, by taking the grayscale mean of the image blocks before and after enhancement and , grayscale variance and the covariance between the two Statistical extraction is performed to construct a comprehensive evaluation model based on brightness consistency, contrast preservation, and structural reconstruction. After normalization and suppression adjustment, an evaluation value representing the structural consistency of the image block is generated. The calculation formula is:

[0079] ;

[0080] Where, and For image blocks Instead of enhancing the average value of image patches, is the covariance of the image block and the enhanced image block, that is, , and The constant is mainly used to avoid division by zero errors during calculation. , ;

[0081] S42. The edge continuity index is constructed by analyzing the difference in edge gradients before and after image block enhancement. First, the gradient amplitude map is extracted from each image block, and the gradient response value of each pixel position in the original image and the enhanced image is recorded respectively. Then, the point-by-point difference of the response value before and after the same position is calculated, and the average is taken over the entire infrared thermal imaging image area to obtain the edge disturbance statistics. The edge continuity after enhancement is calculated using the following formula:

[0082] ;

[0083] Where, is the edge continuity score of the image block, which is used to measure the edge continuity of the image block after enhancement. For image blocks No. The gradient of a pixel represents the edge strength of the pixel. When calculating the gradient, the Sobel operator is used to calculate the gradient in the horizontal and vertical directions. The enhanced image block No. The gradient of pixels, is the number of pixels in the image block to be calculated. In this implementation, N≤10000;

[0084] S43. Calculate the brightness deviation between each pixel in the image block and its neighborhood, and then normalize it to a standard contrast response. This response is determined by the pixel value and the grayscale mean and standard deviation of the image block in which it is located. After performing the above processing on the image block before and after enhancement, calculate the difference between the two, and perform statistical aggregation over the entire block to calculate the local contrast improvement score. The calculation formula is:

[0085] ;

[0086] Where, yes The local contrast improvement score of the image block is used to measure the contrast enhancement degree of the enhanced image in the local area. For image blocks No. The gray value of a pixel, The enhanced image block No. The gray value of a pixel, is an image block The mean value of all pixels in , , is the enhanced image block The mean value of all pixels in ;

[0087] S44. The fusion score takes the three indicators of structural similarity, edge continuity and local contrast as input, generates the overall colony image block quality evaluation value based on the multi-factor weighting mechanism, and feeds it back to the parameter adjustment module. Each sub-indicator is assigned a fixed weight coefficient according to its importance in the task. The linear superposition strategy is adopted in the fusion process, and the scale uniformity between the indicators is considered. The calculation formula is:

[0088] ;

[0089] Where, is the weight coefficient, satisfying .

[0090] S5. Use the structural frequency-curvature coupling to map the thermal imaging image structure, calculate the image local frequency response, regional curvature value, information entropy, and construct a nonlinear coupling mapping function.

[0091] Further, in step S5, as Figure 4 As shown in FIG, the specific steps of infrared thermal imaging image structure frequency curvature coupling mapping are as follows.

[0092] S51. Perform wavelet transform on the original infrared thermal imaging image to extract its sub-band responses in the horizontal, vertical and diagonal directions, which respectively reflect the texture intensity changes of the infrared thermal imaging image in different directions. For each pixel position, count its response amplitude in the three directional sub-bands, obtain its overall frequency response intensity through the amplitude combination strategy, and calculate the local frequency response of the image. , the formula for calculating the high frequency component of each pixel through wavelet transform is:

[0093] ;

[0094] Where, are the wavelet coefficients of the infrared thermal imaging image in the horizontal, vertical and diagonal directions, with initial values ​​of 1, 1 and 2 respectively;

[0095] S52. By obtaining the gradient values ​​of the infrared thermal imaging image in the horizontal and vertical directions, further calculate its second-order derivative, including the lateral change rate, longitudinal change rate and mixed change rate. In the processing process, a small regularization factor is introduced to calculate the local curvature value. , the calculation formula based on the second-order derivative is:

[0096] ;

[0097] Where, It is The local curvature value at a pixel indicates the degree of curvature of the surface in the area where the pixel is located. It is used to detect edges and curved areas in the image. is the gradient of the infrared thermal imaging image in the x direction, that is, the partial derivative, is the gradient of the infrared thermal imaging image in the y direction, is the second-order derivative of the infrared thermal imaging image in the x direction, is the mixed second-order derivative of the infrared thermal imaging image in the xy direction, is the second-order derivative of the infrared thermal imaging image in the y direction, is a small constant, , used to avoid the denominator being zero;

[0098] S53, establish a sliding window with each pixel as the center, count the gray value distribution of all pixels in the window, and calculate its probability distribution, construct an entropy function based on the probability density, and add a minimal bias to calculate the information entropy in the process of probability density construction , the calculation formula based on local grayscale distribution is:

[0099] ;

[0100] Where, Grayscale value The probability distribution of is a constant, and is initially set to 1 to prevent zero from appearing in logarithmic calculations;

[0101] S54. The generation of enhancement weights is based on the three characteristic parameters of structural frequency, local curvature and information entropy, which are integrated through a nonlinear coupling mechanism. First, the scale of the above three types of features is normalized. Then, control factors are allocated according to task requirements to adjust the influence of each parameter on the enhancement weight. Then, control factors are allocated according to task requirements to adjust the influence of each parameter on the enhancement weight. The enhancement parameters are generated through nonlinear coupling mapping. The calculation formula is:

[0102] ;

[0103] Where, is the final enhancement weight.

[0104] Furthermore, in step S5, for the infrared thermal imaging image enhancement model, it is written in Python language and uses the Pytorch framework. The optimizer uses stochastic gradient descent, the initial learning rate is set to 0.01, the momentum is set to 0.9, the weight decay is set to 0.0001, the training batch is set to 64, and the number of training rounds is initially set to 100. The cross-entropy loss function is used to calculate the loss of each iteration, and the model parameters are optimized based on the loss. After training, the structural similarity index is used to evaluate the performance of the model, reflecting the model's ability to preserve details and suppress noise during the enhancement process.

[0105] S6. Merge the infrared thermal imaging image blocks to obtain an overall enhanced infrared thermal imaging image output.

[0106] Further, in step S6, as Figure 5 As shown, Figure 5 The effect of infrared thermal imaging image before enhancement is shown. Figure 6 As shown, Figure 6 The enhanced image effect obtained by processing the infrared thermal imaging image enhancement model is shown.

[0107] The above are only preferred embodiments of the present invention. It should be pointed out that those skilled in the art can make several modifications and improvements without departing from the inventive concept of the present invention, and these all fall within the scope of protection of the present invention.

Claims

1. A method for processing infrared thermal imaging images for painful areas, characterized in that: The following steps are involved: S1. Constructing an initial structural description, collecting infrared thermal imaging images of the pain area, constructing a preliminary abnormal area distribution map based on edge segmentation, and producing a structural vector map; including: annotating regions of interest for each thermal imaging image, wherein the regions of interest include abnormally heated edges, core hot spots, and peripheral transition zones, constructing a preliminary abnormal thermal distribution map based on edge segmentation, and producing a structural vector map of the pain area; S2. Extract the number, position, size, and edge blur of abnormal regions based on the structural vector map to form a structural guidance map. The structural guidance map is constructed by extracting the spatial position information, scale information, and boundary blur index of each abnormal region from the abnormal region distribution map, encoding the multidimensional attribute information into a feature combination for structural modeling, wherein the description unit of each abnormal region includes coordinate information indicating its positional relationship in the infrared thermal imaging image, a scale parameter reflecting its growth range, and an edge blur parameter for measuring the edge clarity. The structural vector set is used to construct the overall structural guidance map. ; S3. Cutting the infrared thermal imaging image into multiple image blocks, and performing enhancement processing on each image block based on its local features and the structure guidance map; including: S31. Based on a preset two-dimensional grid partitioning strategy, the input infrared thermal imaging image is evenly divided in the spatial dimension to form a plurality of non-overlapping sub-image regions. Each sub-image region has a uniform size in the row and column directions. The partitioning result covers the entire infrared thermal imaging image. Each image block is assigned a clear position index identifier and is matched and mapped to the structure guidance map at the regional level. S32. The enhancement processing method adopts an image block enhancement strategy based on feature response guidance. By analyzing the brightness distribution characteristics of the current image block, the structural guidance information, and the spatial position relationship of the current image block in the overall infrared thermal imaging image, the contrast adjustment amplitude and detail retention strength of the image block are dynamically adjusted. The enhancement parameters are determined based on the image statistical characteristics of the current block and the corresponding structural guidance area. Differentiated enhancement operations are applied to each image block in a block-by-block processing manner. S4. Calculate the three indicators of structural similarity, edge continuity, and local contrast for the enhanced image block, calculate the quality score, and adjust the enhancement parameters of the next image block according to the score; S5. Use structural frequency-curvature coupling to map the thermal imaging image structure, calculate the image local frequency response, regional curvature value, information entropy, and construct a nonlinear coupling mapping function; S6. Merge the image blocks to obtain an overall enhanced thermal imaging image output.

2. The method for processing infrared thermal imaging images targeting pain areas according to claim 1, characterized in that: In step S4, the quality score is calculated as follows: S41, calculate the structural similarity SSIM of each image block after enhancement, by taking the grayscale mean of the image blocks before and after enhancement and , grayscale variance and and the covariance between the two Perform statistical extraction and build a comprehensive evaluation model based on brightness consistency, contrast preservation, and structural reconstruction. After normalization and suppression adjustment, an evaluation value representing the structural consistency of the image block is generated. S42. The edge continuity index is constructed by analyzing the difference in edge gradients before and after image block enhancement. First, the gradient amplitude map is extracted from each image block, and the gradient response value of each pixel position in the original image and the enhanced image is recorded respectively. Then, the point-by-point difference of the response value before and after the same position is calculated, and the average is taken over the entire infrared thermal imaging image area to obtain the edge disturbance statistics. The edge continuity after enhancement is calculated using the following formula: ； Where, is the edge continuity score of the image block, For image blocks No. The gradient of pixels, The enhanced image block No. The gradient of pixels, is the number of pixels in the image block to be calculated; S43. Calculate the brightness deviation between each pixel in the image block and its neighborhood, and then normalize it to a standard contrast response. This response is determined by the pixel value and the grayscale mean and standard deviation of the image block in which it is located. After performing the above processing on the image block before and after enhancement, calculate the difference between the two, and perform statistical aggregation over the entire block to calculate the local contrast improvement score. The calculation formula is: ； Where, yes The local contrast improvement score of the image patch, For image blocks No. The gray value of a pixel, The enhanced image block No. The gray value of a pixel, is an image block The mean value of all pixels in , is the enhanced image block The mean value of all pixels in , For image blocks The standard deviation of The enhanced image block The standard deviation of S44. The fusion score takes structural similarity, edge continuity and local contrast as input, generates an overall image block quality evaluation value based on a multi-factor weighting mechanism, and feeds it back to the parameter adjustment module. Each sub-indicator is assigned a fixed weight coefficient according to its importance in the task. A linear superposition strategy is adopted in the fusion process, while considering the scale uniformity between indicators.

3. The method for processing infrared thermal imaging images targeting pain areas according to claim 2, characterized in that: In step S5, the calculation method of the infrared thermal imaging image structure frequency curvature coupling map is: S51. Perform wavelet transform on the original infrared thermal imaging image to extract its sub-band responses in the horizontal, vertical and diagonal directions, which respectively reflect the texture intensity changes of the infrared thermal imaging image in different directions. For each pixel position, count its response amplitude in the three directional sub-bands, obtain its overall frequency response intensity through the amplitude combination strategy, and calculate the local frequency response of the image. ; S52. By obtaining the gradient values ​​of the infrared thermal imaging image in the horizontal and vertical directions, further calculate its second-order derivative, including the lateral change rate, longitudinal change rate and mixed change rate. In the processing process, a small regularization factor is introduced to calculate the local curvature value. ; S53, establish a sliding window with each pixel as the center, count the gray value distribution of all pixels in the window, and calculate its probability distribution, construct an entropy function based on the probability density, and add a minimal bias to calculate the information entropy in the process of probability density construction ; S54. The generation of enhancement weights is based on three characteristic parameters: structural frequency, local curvature and information entropy, and is integrated through a nonlinear coupling mechanism. First, the three characteristic parameters are scale-normalized, and then control factors are allocated according to task requirements to adjust the degree of influence of each parameter on the enhancement weights. Then, control factors are allocated according to task requirements to adjust the degree of influence of each parameter on the enhancement weights, and enhancement parameters are generated through nonlinear coupling mapping.

Images