Pressure damage classification system based on image processing

Abstract

The invention discloses a pressure damage classification system based on image processing, which belongs to the technical field of image processing and comprises a pressure damage image acquisition module, a pressure damage image preprocessing module, a pressure damage classification model construction module and a pressure damage classification module. According to the method, a comprehensive noise characteristic value is obtained based on a pixel variance, an LBP value variance and a distance weight, an adaptive filtering weight is calculated, bilateral filtering and Gaussian filtering are fused for denoising, a detail enhancement factor is generated in combination with a local entropy value and a Sobel gradient amplitude, detail enhancement is carried out, the local contrast is enhanced, and the image quality is remarkably improved; according to the method, multi-scale features are extracted and fused, a space-channel attention mechanism is introduced to obtain a global weight, global enhanced fusion features and local enhanced fusion features are obtained, and comprehensive features are generated through fusion, so that the capability of identifying stress damage in different stages is remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the field of image processing technology, and specifically refers to a pressure injury classification system based on image processing. Background Art

[0002] The pressure injury classification system is an intelligent system that uses advanced image processing technology to study a large number of patients' skin pressure injury image data, explore the characteristics and patterns therein, and accurately classify pressure injuries. It provides medical staff with an objective and accurate basis for injury staging, helps to develop more targeted treatment and care plans, effectively improves the diagnosis and treatment efficiency and quality of pressure injuries, and improves the patient's recovery process.

[0003] However, existing pressure injury classification systems have problems such as unstable pressure injury image acquisition quality, severe background interference and uneven lighting conditions, large differences in noise distribution, unclear injury details and insufficient contrast, which make image feature extraction difficult and thus affect the accuracy of pressure injury classification. In the existing pressure injury classification system, pressure injuries exhibit multi-scale, multi-morphological, blurred boundaries and subtle feature differences at different stages. Traditional single-scale feature extraction methods find it difficult to fully capture these complex features, resulting in insufficient recognition ability of pressure injuries at different stages. Summary of the Invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a pressure injury classification system based on image processing. In view of the problems in the existing pressure injury classification system, such as unstable pressure injury image acquisition quality, serious background interference and uneven lighting conditions, large noise distribution differences, unclear injury details and insufficient contrast, which make image feature extraction difficult and thus affect the accuracy of pressure injury classification, this scheme eliminates background interference by segmenting the pressure injury image; obtains comprehensive noise characteristic values ​​based on pixel variance, LBP value variance and distance weight, evaluates noise intensity, and avoids information loss or residual noise caused by unified denoising; calculates adaptive filtering weights, and fuses bilateral filtering and Gaussian filtering for denoising to enhance image usability and feature distinguishability; combines local entropy value and Sobel gradient amplitude to generate detail enhancement factor, performs detail enhancement, and enhances local contrast. The image quality is improved significantly, thereby improving the accuracy of pressure injury staging. In the existing pressure injury classification system, pressure injuries exhibit multi-scale, multi-morphological, fuzzy boundaries and subtle feature differences in different stages. Traditional single-scale feature extraction methods are difficult to fully capture these complex features, resulting in insufficient recognition of pressure injuries of different stages. This solution extracts multi-scale features and fuses them to enhance the information interaction between features of different scales. The spatial-channel attention mechanism is introduced to obtain global weights, strengthen high-importance features and weighted low-importance features, obtain global enhanced fusion features, obtain local enhanced fusion features through channel grouping and weighting, fuse to generate comprehensive features, classify and output image labels, enhance the model's comprehensive discrimination ability for injuries of different stages, and significantly improve the recognition ability of pressure injuries of different stages.

[0005] The pressure injury classification system based on image processing provided by the present invention includes a pressure injury image acquisition module, a pressure injury image preprocessing module, a pressure injury classification model construction module and a pressure injury classification module;

[0006] The pressure injury image acquisition module acquires historical patient skin pressure injury images and corresponding injury categories;

[0007] The pressure injury image preprocessing module segments the pressure injury image, obtains a comprehensive noise characteristic value based on pixel variance, LBP value variance, and distance weight, evaluates noise intensity, calculates adaptive filtering weights, and performs denoising by fusing bilateral filtering with Gaussian filtering. It generates a detail enhancement factor based on local entropy and Sobel gradient amplitude, performs detail enhancement, and enhances local contrast to obtain a processed image of the skin region.

[0008] The pressure injury classification model module extracts and fuses multi-scale features, introduces a spatial-channel attention mechanism to obtain global weights, strengthens high-importance features and weighted low-importance features, obtains global enhanced fusion features, obtains local enhanced fusion features through channel grouping and weighting, fuses and generates comprehensive features, classifies and outputs image labels;

[0009] The pressure injury classification module collects real-time pressure injury images of the patient's skin, inputs them into the pressure injury classification model after preprocessing, and obtains the corresponding injury category based on the output image label.

[0010] Furthermore, the pressure injury image acquisition module acquires historical patient skin pressure injury images and corresponding injury categories, uses the injury categories as image labels, and annotates the pressure injury images.

[0011] Furthermore, the pressure injury image preprocessing module is provided with a noise intensity assessment unit, an adaptive filtering denoising unit and an enhancement unit, and specifically includes the following contents:

[0012] Noise intensity assessment unit: Perform preliminary filtering on the pressure injury image to obtain a preliminary noise reduction image, and segment it into a skin area image and a background area image according to the optimal segmentation threshold; divide the skin area image into non-overlapping initial blocks, calculate the pixel variance and LBP value variance of each initial block to obtain dual-feature heterogeneity, and set a splitting threshold. If the dual-feature heterogeneity of the initial block is greater than the splitting threshold, the initial block is split into four sub-blocks; for each block, the pixel variance and LBP value variance are normalized, and then combined with the distance weight between the block center coordinate and the skin area image center coordinate to obtain the comprehensive noise characteristic value of each block, and the noise intensity is divided into three categories: low noise intensity block, medium noise intensity block, and high noise intensity block;

[0013] Adaptive filtering denoising unit: For each position coordinate in the skin area image, the adaptive filtering weight is calculated according to the noise intensity of the block where it is located. The results of bilateral filtering and Gaussian filtering are linearly weighted fused using the adaptive filtering weight to obtain a denoised image of the skin area;

[0014] Enhancement unit: In the denoised image of the skin area, a 5×5 pixel neighborhood is taken with each position coordinate as the center, and the local entropy value and Sobel gradient amplitude of the neighborhood are calculated. Then, the detail enhancement factor is generated, and the difference between the preliminary denoised image and the denoised image of the skin area is weighted to obtain a skin area detail enhanced image. Local contrast enhancement is also performed to obtain a skin area processed image.

[0015] Furthermore, the pressure injury classification model module is provided with a multi-scale feature extraction unit, a fusion unit, and an attention feature fusion network unit, specifically including the following contents:

[0016] Multi-scale feature extraction unit: extracts multi-scale features from the skin area processed image, obtains four layers of features, and unifies the number of channels of all features;

[0017] Fusion unit: After upsampling the upper layer features, it adds them to the lower layer features of the same size, and repairs discontinuous edges through convolution to obtain coarse-grained features, medium-grained features, and fine-grained features. Then, through sampling, features of different levels are made to interact at the same size, and combined with BN, ReLU and 3×3 convolution optimization, the corresponding optimized features are obtained.

[0018] Attention feature fusion network unit; includes the following:

[0019] Spatial-channel attention mechanism: uniformly optimize the size and number of channels of features to obtain coarse-grained unified features F1, medium-grained unified features F2, and fine-grained unified features F3, and then concatenate the channels to generate multi-grained unified features. Perform spatial pooling and channel pooling on the multi-grained unified features to obtain spatial attention weights and channel attention weights. After the sizes of both are matched to the multi-grained unified features, they are fused to obtain the global weight.

[0020] Global and local enhancement; perform average pooling on the global weight to obtain the global threshold, perform element-level threshold division on F1, F2, and F3 according to the global threshold to obtain their respective high-importance features and low-importance features. The high-importance features are spliced ​​and enhanced by depth-wise separable convolution and multiplied with the global weight. The low-importance features are spliced ​​and weighted by the weights reconstructed by Softmax. The enhanced high-importance features and the weighted low-importance features are added element by element to obtain the global enhanced fusion features; local enhanced features are obtained based on multi-granularity unified features, and the local enhanced features are evenly divided into 8 groups by channel. Each group generates a weight distribution that matches its own number of channels through 1×1 convolution, which is normalized by Softmax as the channel weight of the group. The channel weight of each group is used to weight the features of the group element by element, and the local enhanced fusion features are obtained after splicing;

[0021] Classify and output; add the global enhanced fusion features and the local enhanced fusion features element by element to obtain the comprehensive features; perform global average pooling on the optimized features and the comprehensive features respectively to obtain 4 feature vectors, which are then concatenated and passed through two fully connected layers and Softmax normalization, and the damage category with the highest probability is output as the image label.

[0022] Furthermore, the pressure injury classification module collects real-time pressure injury images of the patient's skin, inputs them into the pressure injury classification model after preprocessing, obtains the corresponding injury category based on the output image label, and completes the real-time pressure injury classification of the patient's skin.

[0023] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0024] (1) In view of the problems in the existing pressure injury classification system, such as unstable pressure injury image acquisition quality, serious background interference and uneven lighting conditions, large differences in noise distribution, unclear injury details and insufficient contrast, which make image feature extraction difficult and thus affect the accuracy of pressure injury classification, this scheme segments the pressure injury image to eliminate background interference and improve the pertinence and accuracy of feature extraction; obtains the comprehensive noise characteristic value based on pixel variance, LBP value variance and distance weight, evaluates the noise intensity, realizes quantitative evaluation of noise intensity, avoids information loss or residual noise caused by unified denoising, and improves the consistency and controllability of image quality; calculates adaptive filtering weights and fuses bilateral filtering and Gaussian filtering for denoising to enhance image usability and feature distinguishability; combines local entropy value and Sobel gradient amplitude to generate detail enhancement factor, performs detail enhancement, and enhances the model's recognition ability of injuries of different stages; and enhances local contrast to obtain skin area processed image, which significantly improves image quality and thus improves the accuracy of pressure injury classification.

[0025] (2) In view of the fact that pressure injuries in the existing pressure injury classification system exhibit multi-scale, multi-morphological, blurred boundaries and subtle feature differences at different stages, the traditional single-scale feature extraction method is difficult to fully capture these complex features, resulting in insufficient recognition ability of pressure injuries at different stages. This scheme extracts multi-scale features and fuses them to comprehensively cover the details and structural features of the injury and enhance the information interaction between features at different scales; introduces a spatial-channel attention mechanism to obtain global weights, strengthens high-importance features and weighted low-importance features, obtains global enhanced fusion features, strengthens features that contribute greatly to classification, and retains auxiliary information of low-importance features; obtains local enhanced fusion features through channel grouping and weighting, and enhances the model's ability to discriminate injury details; fuses to generate comprehensive features, classifies and outputs image labels, enhances the model's comprehensive discrimination ability for injuries at different stages, and significantly improves the recognition ability of pressure injuries at different stages. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A schematic diagram of the pressure injury classification system based on image processing provided by the present invention;

[0027] Figure 2Schematic diagram of the pressure injury image preprocessing module;

[0028] Figure 3 Schematic diagram of the module for constructing the pressure injury classification model.

[0029] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0031] In the description of the present invention, it should be understood that terms such as "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0032] Example 1, see Figure 1 The pressure injury classification system based on image processing provided by the present invention includes a pressure injury image acquisition module, a pressure injury image preprocessing module, a pressure injury classification model construction module and a pressure injury classification module;

[0033] The pressure injury image acquisition module acquires historical patient skin pressure injury images and corresponding injury categories, and sends the data to the pressure injury image preprocessing module;

[0034] The pressure injury image preprocessing module receives data sent by the pressure injury image acquisition module, segments the pressure injury image, obtains a comprehensive noise characteristic value based on pixel variance, LBP value variance and distance weight, evaluates noise intensity, calculates adaptive filtering weights, and performs denoising by fusing bilateral filtering with Gaussian filtering. It generates a detail enhancement factor based on local entropy value and Sobel gradient amplitude, performs detail enhancement, and enhances local contrast to obtain a skin region processed image, and sends the data to the pressure injury classification model construction module;

[0035] The pressure injury classification model construction module receives data sent by the pressure injury image preprocessing module, extracts and fuses multi-scale features, introduces a spatial-channel attention mechanism to obtain global weights, strengthens high-importance features and weighted low-importance features, obtains global enhanced fusion features, obtains local enhanced fusion features through channel grouping and weighting, fuses to generate comprehensive features, classifies and outputs image labels, and sends the data to the pressure injury classification module;

[0036] The pressure injury classification module receives data sent by the pressure injury classification model construction module, collects real-time pressure injury images of the patient's skin, inputs them into the pressure injury classification model after preprocessing, and obtains the corresponding injury category based on the output image label.

[0037] Example 2, see Figure 1 This embodiment is based on the above embodiment. In the pressure injury image acquisition module, historical patient skin pressure injury images and corresponding injury categories are collected, and the injury categories are used as image labels to annotate the pressure injury images. The injury categories include congestion and redness stage, inflammatory infiltration stage, shallow ulcer stage, necrotic ulcer stage, unclassifiable stage, and deep tissue injury stage.

[0038] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In the pressure injury image preprocessing module, a noise intensity evaluation unit, an adaptive filtering denoising unit, and an enhancement unit are provided. Specifically, the following contents are included:

[0039] Noise intensity assessment unit; pressure injury images come from diverse sources, with inconsistent formats and pixel ranges, and contain irrelevant backgrounds that interfere with feature extraction in the injury area. Furthermore, the noise distribution in pressure injury images is uneven, and unified denoising can easily lead to smoothing of injury details or incomplete denoising of high-noise areas. By dividing the image blocks and calculating the comprehensive noise characteristic value, we accurately divide the low, medium, and high noise intensity blocks, quantify the noise intensity in different areas, and provide a basis for subsequent adaptive denoising, ensuring that high-noise intensity blocks are fully denoised and low-noise intensity blocks retain details to avoid feature loss; uniformly convert the pressure injury images into 8-bit unsigned The image was converted into a grayscale image and the Otsu algorithm was applied to calculate the optimal segmentation threshold. The preliminary denoised image was segmented into a skin region image and a background region image according to the optimal segmentation threshold. The skin region image was divided into non-overlapping initial blocks of 16×16. The pixel variance and LBP value variance of each initial block were calculated to obtain the dual-feature heterogeneity. The average dual-feature heterogeneity of all blocks was used to calculate the optimal segmentation threshold. Set the split threshold If the dual-feature heterogeneity of the initial block is greater than the split threshold, the initial block is split into four 8×8 sub-blocks. For each block, the pixel variance and LBP value variance are normalized, and then combined with the distance weight between the block center coordinate and the skin area image center coordinate to obtain the comprehensive noise characteristic value of each block. According to the size of the comprehensive noise characteristic value, the noise intensity is divided into three categories: low noise intensity block, medium noise intensity block and high noise intensity block. The formula used is as follows:

[0040] ;

[0041] ;

[0042] ;

[0043] ;

[0044] Where, and are the pixel variance and LBP value variance of block A, C A , L A and K A They are the dual feature heterogeneity, comprehensive noise feature value and noise intensity classification mark of block A, K A 0, 1, and 2 indicate that block A is a low noise intensity block, a medium noise intensity block, and a high noise intensity block, respectively; D A is the center coordinate of block A The coordinates of the center of the skin area image The distance weight, R is the diagonal length of the skin area image, and are the maximum pixel variance and the maximum LBP value variance in all blocks, respectively; T1 is the classification threshold for low noise intensity blocks and medium noise intensity blocks; T2 is the classification threshold for medium noise intensity blocks and high noise intensity blocks;

[0045] Adaptive filtering denoising unit; a single filter cannot adapt to the uneven noise in the damaged area, and the filter weight is dynamically adjusted according to the noise intensity classification, combining the advantages of the two filters. While removing noise, it retains the key features of the damage to the greatest extent, retaining effective information for subsequent feature extraction; for each position coordinate in the skin area image , the adaptive filtering weight is calculated according to the noise intensity of the block where it is located, and the results of bilateral filtering and Gaussian filtering are linearly weighted fused through the adaptive filtering weight to obtain the denoised image of the skin area; the formula used is as follows:

[0046] ;

[0047] ;

[0048] Where, is the skin area image at position coordinates The adaptive filter weights, is the skin area image at position coordinates The noise intensity classification identifier of the block to which the location belongs, is the denoised image of the skin area at position coordinates The pixel value at and The skin area images after bilateral filtering and Gaussian filtering are at the position coordinates The pixel value at ;

[0049] Enhancement unit: The characteristics of early pressure injuries are not obvious, or the injury details are weakened after denoising, which makes it difficult for the model to capture the stage differences. The enhancement factor is generated by local entropy and Sobel gradient, and the difference between the denoised image and the initial denoised image is weighted to enhance the injury details and highlight the characteristics of low-contrast injuries, making the subtle differences between different stages easier to be recognized by the model; in the denoised image of the skin area, the coordinates of each position are used to generate the enhancement factor. A 5×5 pixel neighborhood is taken as the center, and the local entropy value and Sobel gradient amplitude of the neighborhood are calculated. The local entropy value and Sobel gradient amplitude are combined to generate a detail enhancement factor. The skin area detail enhanced image is obtained by weighting the difference between the initial denoised image and the skin area denoised image. The limited contrast adaptive histogram equalization algorithm is used to perform local contrast enhancement on the skin area detail enhanced image to obtain the skin area processed image. The formula used is as follows:

[0050] ;

[0051] ;

[0052] ;

[0053] ;

[0054] Where, and The position coordinates are The local entropy value and Sobel gradient amplitude of the neighborhood centered on is , i is the pixel value index, is the probability of pixel value i appearing in the neighborhood, and The position coordinates are The horizontal and vertical gradients of the neighborhood centered at E max and are the maximum local entropy value and the maximum Sobel gradient amplitude in the denoised image of the skin area, respectively. is the denoised image of the skin area at position coordinates The detail enhancement factor, and The skin area detail enhanced image and the preliminary denoised image are respectively at the position coordinates The pixel value at .

[0055] By performing the above operations, the existing pressure injury classification system has problems such as unstable pressure injury image acquisition quality, severe background interference, uneven lighting conditions, large noise distribution differences, unclear injury details, and insufficient contrast, which make image feature extraction difficult and thus affect the accuracy of pressure injury classification. This solution segments the pressure injury image to eliminate background interference and improve the pertinence and accuracy of feature extraction. It also obtains a comprehensive noise characteristic value based on pixel variance, LBP value variance, and distance weight, evaluates the noise intensity, and achieves a quantitative assessment of the noise intensity, avoiding information loss or residual noise caused by unified denoising and improving the consistency and controllability of image quality. It calculates adaptive filtering weights and combines bilateral filtering with Gaussian filtering for denoising, enhancing image usability and feature distinguishability. It combines local entropy and Sobel gradient amplitude to generate a detail enhancement factor for detail enhancement, enhancing the model's ability to recognize injuries of different stages. It also enhances local contrast to obtain a processed image of the skin region, significantly improving image quality and thus improving the accuracy of pressure injury staging.

[0056] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In the module for constructing the pressure injury classification model, a multi-scale feature extraction unit, a fusion unit, and an attention feature fusion network unit are provided. Specifically, the following contents are included:

[0057] Multi-scale feature extraction unit; The characteristics of pressure injuries are multi-scale, and a single-scale feature cannot cover the key information of different stages. ResNet50 is used to extract multi-scale features from the bottom layer to the top layer. The residual structure of ResNet50 avoids gradient disappearance and can effectively capture the characteristics of different stages, providing a comprehensive feature basis for subsequent fusion; ResNet50 is used to extract multi-scale features from the skin area processing image, and the bottom-level feature B1, middle-level feature B2, high-level feature B3 and top-level feature B4 are obtained based on the output of the Conv2_x, Conv3_x, Conv4_x and Conv5_x residual block group, and the number of channels of all features is unified to 256 through 1×1 convolution; Among them, the size of the skin area processing image is H×W, and the sizes of B1, B2, B3 and B4 are respectively 、 、 and ;

[0058] Fusion unit; high- and low-level features exist in isolation and cannot reflect the local details of the damage and the overall structural association. Through fusion, the coarse-grained features contain more details, and the fine-grained features contain more global information, thereby improving the representation ability of features for different stages; after upsampling the upper-level features, they are added to the lower-level features of the same size, and the discontinuous edges of the upsampling are repaired by 3×3 convolution to obtain coarse-grained features M1, medium-grained features M2, and fine-grained features M3; then, through sampling, features of different levels are made to interact at the same size, and combined with BN, ReLU, and 3×3 convolution optimization, the corresponding optimized features are obtained. The optimized features include coarse-grained optimized features. , medium-granularity optimization features and fine-grained optimization features ; The formula used is as follows:

[0059] ;

[0060] ;

[0061] ;

[0062] ;

[0063] ;

[0064] ;

[0065] Where, It uses bilinear interpolation algorithm to upsample the input features to twice the original size. is a 3×3 convolution, is an element-wise addition operator, and the sizes of M1, M2, and M3 are 、 and , It uses a 3×3 convolution with a stride of 2 to downsample the input features to twice the original size. It uses bilinear interpolation algorithm to upsample the input features to 4 times the original size. is the rectified linear unit activation function, is batch normalization, 、 and The sizes are 、 and ;

[0066] Attention feature fusion network unit; includes the following:

[0067] Spatial-channel attention mechanism: The feature sizes and channel numbers of different granularities are inconsistent, and the attention mechanism calculation cannot be performed directly, resulting in fusion deviation. Unifying the feature size and channel number ensures that the attention mechanism can act evenly on all granular features to avoid neglecting certain features due to format differences. In addition, the features contain irrelevant information, which interferes with the model's attention to the core features of the damage. Spatial attention is used to highlight the key location of the damage, and channel attention is used to highlight the key feature channels. The two are fused into a global weight, so that the model focuses on the key information of the classification and reduces the interference of irrelevant information. Upsampling 2 times, Downsample by 2 times size, and transform it into 、 and The number of channels is unified to 64, and the coarse-grained unified feature F1, the medium-grained unified feature F2 and the fine-grained unified feature F3 are obtained, and the channels are spliced ​​to generate the multi-grained unified feature ; Perform spatial pooling and channel pooling on the multi-granularity unified feature F respectively to obtain spatial attention weight and channel attention weight. After dimensional expansion to make the sizes of both match the multi-granularity unified feature F, multiply and fuse them element by element to obtain the global weight; the formula used is as follows:

[0068] ;

[0069] ;

[0070] ;

[0071] Where W s and W c They are spatial attention weight and channel attention weight, W b is the global weight, and They are spatial maximum pooling and spatial average pooling, and They are channel maximum pooling and channel average pooling, is the Sigmoid activation function, is a 3×3 depthwise separable convolution, It is channel splicing, is a multi-layer perceptron, is the element-wise multiplication operator, is the dimension expansion function, and the sizes of F1, F2, F3 and F are all ;

[0072] Global and local enhancement: High-importance features are not enhanced, and the auxiliary information of low-importance features is ignored, resulting in insufficient feature integrity. Based on the global threshold, high-importance and low-importance features are divided, high-importance features are enhanced, and low-importance features are weighted, which not only highlights the staging features but also retains auxiliary information, thereby improving the feature's ability to distinguish complex lesions. In addition, the importance of different channels within the same granularity feature is uneven, and some key channels are submerged, affecting the discrimination of local details. After grouping the features, the key channels within the group are weighted by channel weights to enhance the discriminability of local features and improve the model's ability to recognize subtle staging. The global weights are averaged and pooled to obtain the global threshold W. th , according to W th Perform element-level threshold division on F1, F2, and F3 to obtain their respective high-importance features and low-importance features. The high-importance features are concatenated and enhanced by depthwise separable convolution and multiplied with the global weight. The low-importance features are concatenated and weighted by Softmax reconstruction. The enhanced high-importance features and the weighted low-importance features are added element by element to obtain the global enhanced fusion feature F. ef ; and obtain the local enhanced feature F based on the multi-granularity unified feature F e , F e The channels are evenly divided into 8 groups. Each group generates a weight distribution that matches its own channel number through 1×1 convolution. After Softmax normalization, it is used as the channel weight of the group. The channel weight of each group is used to weight the features of the group element by element. After splicing, the local enhanced fusion feature F is obtained. wf ; The formula used is as follows:

[0073] ;

[0074] ;

[0075] ;

[0076] ;

[0077] ;

[0078] Where, 、 and are the high-importance features in F1, F2, and F3, respectively. 、 and are the low-importance features in F1, F2, and F3 respectively; F h is the enhanced high-importance feature, is the weighted low-importance feature, β is the reconstruction weight of the low-importance feature, is global average pooling, and F e The first and eighth grouping features of is the normalized exponential function;

[0079] Classify and output; global and local enhanced features have their own focuses, and using them alone cannot fully characterize the damage characteristics. The comprehensive features contain both the overall morphology and local details of the damage, providing a more comprehensive basis for classification; add the global enhanced fusion features and the local enhanced fusion features element by element to obtain the comprehensive feature F all ;right 、 、 and F all Global average pooling is performed separately to obtain four feature vectors, which are concatenated and passed through two fully connected layers and Softmax normalization to obtain the probability of each damage category. The damage category with the highest probability is output as the image label.

[0080] By performing the above operations, in view of the fact that in the existing pressure injury classification system, pressure injuries exhibit multi-scale, multi-morphological, blurred boundaries and subtle feature differences in different stages, the traditional single-scale feature extraction method is unable to fully capture these complex features, resulting in insufficient recognition ability of pressure injuries of different stages. This solution extracts and fuses multi-scale features to comprehensively cover the details and structural characteristics of the injury and enhance the information interaction between features of different scales; introduces a spatial-channel attention mechanism to obtain global weights, strengthens high-importance features and weighted low-importance features, obtains globally enhanced fusion features, strengthens features that contribute greatly to classification, and retains auxiliary information of low-importance features; obtains locally enhanced fusion features through channel grouping and weighting, and enhances the model's ability to discriminate injury details; fuses to generate comprehensive features, classifies and outputs image labels, enhances the model's comprehensive discrimination ability for injuries of different stages, and significantly improves the recognition ability of pressure injuries of different stages.

[0081] Example 5, see Figure 1 This embodiment is based on the above embodiment. In the pressure injury classification module, real-time pressure injury images of the patient's skin are collected and input into the pressure injury classification model after preprocessing. According to the output image label, the corresponding injury category is obtained to complete the real-time pressure injury classification of the patient's skin.

[0082] Example 6, see Figure 1This embodiment is based on the above embodiment and adopts the "Pressure Ulcer Image Dataset" released by Brno University of Technology in 2021. The dataset contains standardized pressure injury images certified by the International Wound Care Association IWCA, and contains a total of 1,860 pressure injury images, covering 6 types of injury stages, including: congestion and rosy stage: 320 images, inflammatory infiltration stage: 350 images, shallow ulcer stage: 380 images, necrotic ulcer stage: 310 images, unstageable: 280 images, deep tissue injury stage: 220 images; after preprocessing the dataset, the dataset is divided into training set: validation set: test set = 7:1:2 to construct a pressure injury classification model; the classification ability of the pressure injury classification model is evaluated by the test set, with an accuracy rate of 95.7%. In order to verify the clinical practicality of the model, 200 real clinical real-time pressure injury images were collected based on Zhejiang Cancer Hospital and Hangzhou Gongshu District People's Hospital of Integrated Traditional Chinese and Western Medicine, with an accuracy rate of 92.5%.

[0083] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0084] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0085] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. An image processing-based pressure injury classification system, characterized by: It includes a pressure injury image acquisition module, a pressure injury image preprocessing module, a pressure injury classification model building module and a pressure injury classification module; The pressure injury image acquisition module acquires historical patient skin pressure injury images and corresponding injury categories, uses the injury categories as image labels, and annotates the pressure injury images; The pressure injury image preprocessing module segments the pressure injury image, obtains a comprehensive noise characteristic value based on pixel variance, LBP value variance, and distance weight, evaluates noise intensity, calculates adaptive filtering weights, and performs denoising by fusing bilateral filtering with Gaussian filtering. It generates a detail enhancement factor based on local entropy and Sobel gradient amplitude, performs detail enhancement, and enhances local contrast to obtain a processed image of the skin region. The pressure injury classification model module extracts and fuses multi-scale features, introduces a spatial-channel attention mechanism to obtain global weights, strengthens high-importance features and weighted low-importance features, obtains global enhanced fusion features, obtains local enhanced fusion features through channel grouping and weighting, fuses and generates comprehensive features, classifies and outputs image labels; The pressure injury classification module collects real-time pressure injury images of the patient's skin, inputs them into the pressure injury classification model after preprocessing, and obtains the corresponding injury category based on the output image label.

2. The pressure injury classification system based on image processing according to claim 1, characterized in that: The pressure injury image preprocessing module is provided with a noise intensity assessment unit, an adaptive filtering denoising unit, and an enhancement unit, and specifically includes the following contents: Noise intensity assessment unit: Perform preliminary filtering on the pressure injury image to obtain a preliminary noise reduction image, and segment it into a skin area image and a background area image according to the optimal segmentation threshold; divide the skin area image into non-overlapping initial blocks, calculate the pixel variance and LBP value variance of each initial block to obtain dual-feature heterogeneity, and set a splitting threshold. If the dual-feature heterogeneity of the initial block is greater than the splitting threshold, the initial block is split into four sub-blocks; for each block, the pixel variance and LBP value variance are normalized, and then combined with the distance weight between the block center coordinate and the skin area image center coordinate to obtain the comprehensive noise characteristic value of each block, and the noise intensity is divided into three categories: low noise intensity block, medium noise intensity block, and high noise intensity block; Adaptive filtering denoising unit; Enhancement unit.

3. The pressure injury classification system based on image processing according to claim 2, characterized in that: The adaptive filtering denoising unit calculates the adaptive filtering weight for each position coordinate in the skin area image according to the noise intensity of the block where it is located, and linearly weightedly fuses the results of bilateral filtering and Gaussian filtering through the adaptive filtering weight to obtain a denoised image of the skin area.

4. The pressure injury classification system based on image processing according to claim 3, characterized in that: The enhancement unit takes a 5×5 pixel neighborhood centered at each position coordinate in the skin area denoised image, calculates the local entropy value and Sobel gradient amplitude of the neighborhood, and then combines them to generate a detail enhancement factor. The difference between the preliminary denoised image and the skin area denoised image is weighted to obtain a skin area detail enhanced image, and local contrast enhancement is performed to obtain a skin area processed image.

5. The pressure injury classification system based on image processing according to claim 4, characterized in that: The pressure injury classification model module is equipped with a multi-scale feature extraction unit, a fusion unit, and an attention feature fusion network unit, and specifically includes the following contents: Multi-scale feature extraction unit: extracts multi-scale features from the skin area processed image, obtains four layers of features, and unifies the number of channels of all features; Fusion unit: After upsampling the upper layer features, it adds them to the lower layer features of the same size, and repairs discontinuous edges through convolution to obtain coarse-grained features, medium-grained features, and fine-grained features. Then, through sampling, features of different levels are made to interact at the same size, and combined with BN, ReLU and 3×3 convolution optimization, the corresponding optimized features are obtained. Attention feature fusion network unit.

6. The pressure injury classification system based on image processing according to claim 5, characterized in that: The attention feature fusion network unit includes the following contents: Spatial-channel attention mechanism; uniformly optimize the size and number of channels of features to obtain coarse-grained unified features F1, medium-grained unified features F2, and fine-grained unified features F3, and then splice channels to generate multi-grained unified features; Perform spatial pooling and channel pooling on the multi-granularity unified features respectively to obtain spatial attention weights and channel attention weights. After the sizes of both are matched with the multi-granularity unified features, the global weights are fused. Global and local enhancement; Classify and output.

7. The pressure injury classification system based on image processing according to claim 6, characterized in that: The global and local enhancement is to perform average pooling on the global weight to obtain a global threshold, perform element-level threshold division on F1, F2 and F3 according to the global threshold to obtain their respective high-importance features and low-importance features, and then concatenate the high-importance features and strengthen them through depthwise separable convolution and multiply them with the global weight. After concatenating the low-importance features, they are weighted by the weights reconstructed by Softmax, and the strengthened high-importance features and the weighted low-importance features are added element by element to obtain the globally enhanced fusion features. Based on the multi-granularity unified features, local enhanced features are obtained and evenly divided into 8 groups according to channels. Each group generates a weight distribution that matches its own number of channels through 1×1 convolution. After Softmax normalization, it is used as the channel weight of the group. The channel weight of each group is used to weight the features of this group element by element, and the local enhanced fusion features are obtained after splicing.

8. The pressure injury classification system based on image processing according to claim 7, characterized in that: The classification and output is to add the global enhanced fusion features and the local enhanced fusion features element by element to obtain a comprehensive feature; Global average pooling is performed on the optimized features and comprehensive features respectively to obtain four feature vectors. After concatenation, they are passed through two fully connected layers and Softmax normalization, and the damage category with the highest probability is output as the image label.

Images