Spray bottle medicine recommendation method and system based on image processing

Abstract

The invention discloses a spray bottle medicine recommendation method and system based on image processing, and relates to the technical field of image processing. According to the spray bottle medicine recommendation method based on image processing, wound image data are obtained, a wound image structure distinguishing index is analyzed, and then whether the wound image structure distinguishing index is lower than a preset threshold value or not is judged; inputting the wound image data into a pre-trained image enhancement model for enhancement analysis if the wound image data is lower than the preset threshold value, generating wound enhanced image data, and analyzing a wound image confidence index of the wound enhanced image data; and then, based on a pre-trained wound surface identification model, performing wound response index analysis on the enhanced image data, and synthesizing the wound response index with a wound surface image confidence index to obtain a wound condition perception intensity index. Therefore, fusion evaluation of identification accuracy and image credibility is realized, and the reliability of injury identification results and the scientificity of drug recommendation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method and system for recommending medication for a spray bottle based on image processing. Background Art

[0002] Spray bottles, as a basic drug delivery tool, are widely used in the treatment of various skin wounds. Patients or caregivers judge the wound condition based on visual inspection and spray the liquid medicine. However, due to objective factors such as the complex wound morphology, uneven distribution of injuries, and variable lighting conditions, it is difficult to achieve precise medication control based on manual observation alone, which can easily lead to insufficient or excessive spraying, affecting the treatment effect.

[0003] In recent years, image recognition and intelligent processing technologies have been gradually introduced into medical assistive devices. Some spray bottles can perform partition recognition or state classification of wound images to assist in spraying strategies. However, existing methods generally have the following problems: First, the current system generally assumes that the collected images have sufficient clarity and structural recognizability, and does not analyze and evaluate the structural resolution ability of the image itself, resulting in a significant increase in system recognition errors under blurred, reflective, and low-contrast image conditions; second, existing devices lack automatic enhancement and quality compensation mechanisms when the image clarity is insufficient, making it difficult to restore recognizable images for subsequent processing in scenarios with poor image quality, which makes it difficult to control the spray dosage.

[0004] The limitations of existing technologies include at least the following problems: existing technologies lack a linkage mechanism between image quality analysis results and image enhancement processing and wound identification, resulting in the inability to dynamically adjust the processing path when faced with low-quality image input, which seriously affects the recognition accuracy and the reliability of medication decisions. In actual applications, wound images are often limited by factors such as shooting angle, lighting conditions, reflective interference or image blur, resulting in structural instability and numerical drift in the recognition results. Existing technologies also do not establish a linkage relationship between image structure quality and recognition results. As a result, even if the input image quality is poor and the model output credibility is low, the results are still directly used for spray volume decision-making and medication recommendations, which easily leads to the risk that the medication dosage cannot fully match the actual wound needs, making the entire recommendation strategy susceptible to fluctuations in the front-end input quality, insufficient reliability, and low decision robustness, making it difficult to meet the dual requirements of image credibility recognition and risk prevention and control in actual medical applications. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a spray bottle medication recommendation method and system based on image processing, which solves the problem that the existing technology lacks a linkage mechanism between image quality assessment and wound recognition results, resulting in inaccurate medication recommendations.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a spray bottle medication recommendation method based on image processing, comprising the following steps: obtaining wound image data of a patient to be medicated, and analyzing the wound image structure resolution index of the patient to be medicated; judging whether the wound image structure resolution index of the patient to be medicated is lower than a preset wound image structure resolution index threshold; if the wound image structure resolution index of the patient to be medicated is lower than the preset wound image structure resolution index threshold, inputting the wound image data of the patient to be medicated into a pre-trained image enhancement model for enhancement analysis to obtain wound enhanced image data of the patient to be medicated, and analyzing the wound image confidence index of the patient to be medicated; analyzing the wound enhanced image data of the patient to be medicated based on the pre-trained wound recognition model, and performing a comprehensive analysis with the wound image confidence index to obtain the trauma perception intensity index of the patient to be medicated; and taking preset medication recommendation measures for the patient to be medicated based on the trauma perception intensity index.

[0007] Furthermore, the wound image data specifically refers to the pixel value and two-dimensional coordinates of each pixel point in the wound image. The specific steps for analyzing the wound image structure resolution index of the patient to be treated are as follows: based on the wound image data of the patient to be treated, analyzing the image evaluation set of the patient to be treated, including the wound image spectral segregation index, the wound boundary fluctuation abnormality index, and the wound reflective fragmentation index; performing a comprehensive analysis on the image evaluation set of the patient to be treated to obtain the wound image structure resolution index of the patient to be treated.

[0008] Furthermore, the specific formula for calculating the structural resolution index of the wound surface image of the patient to be treated is as follows: Among them, TqD is the structural resolution index of the wound image of the patient to be treated, PyZ is the spectrum clustering index of the wound image of the patient to be treated, α1 is the clustering coefficient stored in the database, BdC is the abnormal boundary fluctuation index of the wound of the patient to be treated, α2 is the boundary coefficient stored in the database, FgP is the reflective fragmentation index of the wound of the patient to be treated, α3 is the fragmentation coefficient stored in the database, and α4 is the interaction coefficient stored in the database.

[0009] Furthermore, the image enhancement model is specifically a conditional generative adversarial network, including a generator for generating wound-enhanced image data. The generator includes an input encoding layer, a residual modeling layer, a structural attention enhancement layer, an image decoding and reconstruction layer, and a discriminator for distinguishing the generated wound-enhanced image from the real wound image.

[0010] Furthermore, the pre-training steps of the conditional generative adversarial network are as follows: obtain several groups of wound image samples, including damaged wound images and corresponding clear wound images, and divide them into wound image training sets and wound image verification sets; initialize the conditional generative adversarial network, including parameter initialization of the weights and biases of the generator and discriminator; based on the wound image training set, set the training rounds, and perform training loop processing, that is, the generator and discriminator are updated in turn in each training loop; after each training loop, perform evaluation analysis based on the wound image verification set, and adjust the parameters of the conditional generative adversarial network based on the evaluation analysis until the image enhancement result output by the conditional generative adversarial network meets the preset image quality standard.

[0011] Furthermore, the wound enhancement image data specifically includes the enhanced pixel value and enhanced two-dimensional coordinates of each enhanced pixel point in the wound image. The specific steps for analyzing the wound response index of the patient to be treated are as follows: the wound enhancement image data of the patient to be treated is input into the pre-trained wound recognition model for feature extraction and analysis to obtain a wound status feature set of the patient to be treated, including the edge regeneration tension fracture index, heterochromatic cluster offset index, reflection path continuity loss index, and structural halo interference index of each wound area; the wound assessment set of the patient to be treated is comprehensively analyzed to obtain a comprehensive wound status index of each wound area of ​​the patient to be treated, and the mean is processed to obtain the wound response index of the patient to be treated.

[0012] Furthermore, the wound recognition model is specifically an image structure neural network, including an input layer, a structure coding layer, a region segmentation layer, and a feature output layer. The specific steps for obtaining a wound state feature set of the patient to be treated are as follows: in the input layer of the image structure neural network, the wound enhanced image data of the patient to be treated is received and preprocessed; in the structure coding layer of the image structure neural network, the preprocessed wound enhanced image data of the patient to be treated is encoded to obtain the wound structure perception vector and structure feature map tensor of the patient to be treated; in the region segmentation layer of the image structure neural network, the wound structure perception vector and structure feature map tensor of the patient to be treated are segmented to obtain the structure feature vectors of several wound areas of the patient to be treated; in the feature output layer of the image structure neural network, the structure feature vector of each wound area of ​​the patient to be treated is extracted and processed to obtain the edge regeneration tension fracture index, heterochromatic cluster offset index, reflection path continuity loss index, and structure halo interference index of each wound area of ​​the patient to be treated.

[0013] Furthermore, the specific formula for calculating the injury perception intensity index of the patient to be treated is as follows: SgZ = ln(1+η1*CxY)*exp(-η2*TzX)*[1+tanh(η3*CxY*TzX)]; wherein SgZ is the injury perception intensity index of the patient to be treated, CxY is the trauma response index of the patient to be treated, η1 is the trauma response coefficient stored in the database, TzX is the wound image confidence index of the patient to be treated, η2 is the confidence coefficient stored in the database, and η3 is the synergy coefficient stored in the database.

[0014] Furthermore, the specific steps for taking preset medication recommendation measures for patients to be medicated based on the sadness perception intensity index are as follows: comparing and analyzing the sadness perception intensity index of the patient to be medicated with the preset sadness perception intensity index threshold interval; if the sadness perception intensity index of the patient to be medicated is lower than the lower limit of the preset sadness perception intensity index threshold interval, it is marked as low risk, and the first medication recommendation measure is taken; if the sadness perception intensity index of the patient to be medicated is within the preset sadness perception intensity index threshold interval, it is marked as medium risk, and the second medication recommendation measure is taken; if the sadness perception intensity index of the patient to be medicated is higher than the upper limit of the preset sadness perception intensity index threshold interval, it is marked as high risk, and the third medication recommendation measure is taken.

[0015] A spray bottle medication recommendation system based on image processing includes: an image structure resolution module for acquiring wound image data of a patient to be medicated and analyzing a wound image structure resolution index of the patient to be medicated; a resolution judgment module for judging whether the wound image structure resolution index of the patient to be medicated is lower than a preset wound image structure resolution index threshold; an image confidence analysis module for inputting the wound image data of the patient to be medicated into a pre-trained image enhancement model for enhancement analysis if the wound image structure resolution index of the patient to be medicated is lower than the preset wound image structure resolution index threshold, thereby obtaining enhanced wound image data of the patient to be medicated and analyzing a wound image confidence index of the patient to be medicated; a wound recognition analysis module for analyzing the wound enhanced image data of the patient to be medicated based on a pre-trained wound recognition model, and performing a comprehensive analysis with the wound image confidence index to obtain a trauma perception intensity index of the patient to be medicated; and a medication feedback module for taking preset medication recommendation measures for the patient to be medicated based on the trauma perception intensity index.

[0016] The present invention has the following beneficial effects:

[0017] (1) This spray bottle medication recommendation method based on image processing introduces a linkage analysis mechanism of wound image confidence index and trauma response index. When generating the injury perception intensity index, it fully considers the structural credibility of the image itself and avoids directly using low-quality image recognition results when the image is blurred or reflective. This achieves a fusion evaluation of recognition accuracy and image credibility, effectively avoiding the risk of misjudgment caused by problems such as image blur and reflection. The wound image confidence index has the ability to comprehensively judge the clarity, edge contour integrity and structural distribution characteristics of the input image, and forms an effective constraint on the recognition result in the subsequent evaluation process, so that a high-level response is triggered only when the input image meets the premise that the structure can be interpreted, thereby improving the stability and controllability of the overall output of the model, and thereby improving the reliability of the injury recognition results and the scientific nature of the medication recommendation.

[0018] (2) This spray bottle medication recommendation method based on image processing constructs a structural perception-guided image structure neural network model, proposes four types of risk features for wound area division and multi-dimensional structural status assessment, and gradually extracts and models them based on the input layer, structure coding layer, region segmentation layer and feature output layer. In the structure coding layer, different receptive field convolution kernels are used to extract multi-scale structural disturbance information, and weighted fusion is performed through the SE channel attention mechanism to finally generate a structural perception vector. The region segmentation layer combines channel response adjustment and mask position mapping to extract the structural feature vector of each wound area. In the feature output layer, each index is output after being modeled by the corresponding sub-network, thereby realizing automatic perception of the risk pattern of the wound area, fine identification of structural disturbance and index output, thereby laying a scientific foundation for injury grade judgment and quantitative medication.

[0019] (3) The image processing-based spray bottle medication recommendation method, by constructing a multi-channel recognition model based on the wound surface area structure, can output the edge regeneration tension fracture index, the heterochromatic cluster offset index, the reflection path continuity loss index and the structural halo interference index, and calculate the trauma response index based on this. Then, based on the injury perception intensity index generated by the index combined with the image confidence index, the wound surface is divided into mild, moderate and severe levels, and the PLC controller is automatically called to trigger differentiated spray action instructions, corresponding to single, double and continuous spraying schemes respectively. The voice prompt and image synchronization mechanism are combined to achieve intelligent response, thereby improving the automatic recognition ability and dynamic response ability of the spray bottle, and also realizing the direct drive of the spraying behavior by the structural risk characteristics, thereby establishing a full-link closed-loop control path from image structure perception, state assessment to control execution.

[0020] (4) The spray bottle medication recommendation system based on image processing adopts a modular design to achieve a closed-loop automated process of image acquisition, quality assessment, enhancement, recognition and medication feedback. The system calculates the wound image structure resolution index in real time through the image structure resolution module, accurately quantifies the quality status of the input image, and the resolution judgment module automatically decides whether to start image enhancement based on the threshold, effectively avoiding unnecessary waste of computing resources. The image confidence analysis module combines the pre-trained image enhancement model to dynamically improve the structural information and detail clarity of low-quality images, enhancing the generalization performance of the model. Then, the wound recognition analysis module achieves a high-precision assessment of the injury perception intensity index by fusing the trauma response index and the image confidence index, providing a scientific basis for precise medication. The medication feedback module automatically controls the spray volume and mode based on the evaluation results, supports multi-level risk stratification processing, and meets the personalized treatment needs of different injuries, thereby making the overall system architecture clearly layered and functionally coordinated, with both real-time, accuracy and intelligence, and significantly improving the intelligent control level and clinical efficacy of spray medication.

[0021] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flow chart of a spray bottle medication recommendation method based on image processing according to the present invention.

[0023] Figure 2 This is a flowchart of the specific steps of analyzing the structural resolution index of the wound surface image of a patient to be medicated in a spray bottle medication recommendation method based on image processing of the present invention.

[0024] Figure 3 This is a schematic diagram of a wound area sequence of edge regeneration tension fracture index of a patient to be medicated in a spray bottle medication recommendation method based on image processing of the present invention.

[0025] Figure 4 This is a schematic diagram of a sequence of heterochromatic cluster shift index wound surface areas of a patient to be medicated in an image processing-based spray bottle medication recommendation method of the present invention.

[0026] Figure 5 This is a schematic diagram of a wound area sequence of a reflection path continuity loss index of a patient to be medicated in an image processing-based spray bottle medication recommendation method of the present invention.

[0027] Figure 6 This is a schematic diagram of a sequence of structural halo interference index wound surface areas of a patient to be medicated in a spray bottle medication recommendation method based on image processing of the present invention.

[0028] Figure 7 This is a block diagram of the spray bottle medication recommendation system based on image processing in the present invention. DETAILED DESCRIPTION

[0029] See also Figure 1 , an embodiment of the present invention provides a technical solution: a spray bottle medication recommendation method based on image processing, comprising the following steps: obtaining wound image data of a patient to be medicated, and analyzing a structural resolution index of the wound image of the patient to be medicated; judging whether the structural resolution index of the wound image of the patient to be medicated is lower than a preset wound image structural resolution index threshold; if the structural resolution index of the wound image of the patient to be medicated is lower than the preset wound image structural resolution index threshold, inputting the wound image data of the patient to be medicated into a pre-trained image enhancement model for enhancement analysis to obtain enhanced image data of the wound of the patient to be medicated, and analyzing a confidence index of the wound image of the patient to be medicated (consistent with the specific logic of analyzing the structural resolution index of the wound image of the patient to be medicated, i.e., The wound image structure resolution index of the patient to be treated is analyzed and marked as the wound image confidence index); based on the pre-trained wound recognition model, the wound enhanced image data of the patient to be treated is analyzed, and the trauma response index of the patient to be treated is analyzed (it should be noted here that if the wound image structure resolution index of the patient to be treated is not lower than the preset wound image structure resolution index threshold, then the wound image data of the patient to be treated is input here), and the wound image confidence index (it should be noted here that the wound image structure resolution index analyzed based on the wound image data of the patient to be treated is marked as the wound image confidence index) is comprehensively analyzed to obtain the injury perception intensity index of the patient to be treated; based on the injury perception intensity index, the preset medication recommendation measures are taken for the patient to be treated.

[0030] The specific formula for calculating the injury perception intensity index of the patient to be treated is as follows: SgZ = ln(1+η1*CxY)*exp(-η2*TzX)*[1+tanh(η3*CxY*TzX)]; wherein SgZ is the injury perception intensity index of the patient to be treated, CxY is the trauma response index of the patient to be treated, η1 is the trauma response coefficient stored in the database, TzX is the wound image confidence index of the patient to be treated, η2 is the confidence coefficient stored in the database, and η3 is the synergy coefficient stored in the database.

[0031] What needs to be explained is that the specific form of the tanh function is: Here, e is a natural constant and can be 2.71 in this embodiment, with a domain of (-∞, +∞) and a range of (-1, +1).

[0032] η1, η2, and η3 can be obtained through the following steps: using historical data, combined with the trauma response index and wound image confidence index, statistical regression analysis is performed to quantify the specific impact of each factor on the injury perception intensity index, thereby fitting the initial weight value. Secondly, different value intervals of the above two variables are set. While keeping other parameters unchanged, their values ​​are gradually adjusted and the change range of the injury perception intensity index is calculated to test the impact trend of each parameter perturbation on the output results, thereby verifying the stability and applicability of the weight setting on the model output, and accordingly revising the value of the initial weight value.

[0033] Specifically, if Figure 2 As shown, the wound image data specifically includes the pixel value and two-dimensional coordinates of each pixel point in the wound image. The specific steps of analyzing the wound image structure resolution index of the patient to be treated are as follows: based on the wound image data of the patient to be treated, the image evaluation set of the patient to be treated is analyzed, including the wound image spectral segregation index, the wound boundary fluctuation abnormality index (which measures the stability of the edge contour of the wound image), and the wound reflective fragmentation index; the image evaluation set of the patient to be treated is comprehensively analyzed to obtain the wound image structure resolution index of the patient to be treated.

[0034] The specific steps for analyzing the image evaluation set of patients to be treated are as follows:

[0035] The pixel value of each pixel point in the wound surface image of the patient to be treated is gray-scale processed to obtain the gray-scale pixel value of each pixel point in the wound surface image of the patient to be treated, and a fast Fourier transform is performed to obtain a frequency domain amplitude map of the wound surface image of the patient to be treated, including the amplitude values ​​of several frequency coordinate points, and each frequency coordinate point in the frequency domain amplitude map of the wound surface image of the patient to be treated is traversed in turn, and the square of the amplitude value corresponding to the frequency coordinate point is calculated as the amplitude energy value of the point; the central frequency coordinate point of the frequency domain amplitude map is used as a reference, and the Euclidean distance between each frequency coordinate point and the center position is calculated, and the distance is used as the frequency domain radius index; according to the set frequency coordinate point Domain partitioning rules divide the frequency coordinate points into a low-frequency region and a high-frequency region, wherein the low-frequency region is the set of all coordinate points whose frequency domain radius from the center position is less than 25% of the total radius, and the high-frequency region is the set of all coordinate points whose frequency domain radius is greater than 75% of the total radius. In the low-frequency region, the sum of the amplitude energy values ​​corresponding to all frequency coordinate points is counted as the total energy value of the low-frequency region. In the entire frequency domain, the sum of the amplitude energy values ​​of all frequency coordinate points is counted as the total energy value of the full frequency domain. By calculating the ratio of the total energy value of the low-frequency region to the total energy value of the full frequency domain, the spectrum segregation index of the wound image of the patient to be treated with medication is obtained;

[0036] Reading the grayscale pixel value of each pixel in the wound image of the patient to be treated, and performing edge detection processing (i.e., extracting the main edge structure information of the image based on the Canny edge detection operator) to obtain a number of edge pixel points in the wound image of the patient to be treated, in the edge detection process, respectively calculating the grayscale gradient value of each pixel in the wound image in the horizontal and vertical directions based on the Sobel operator (the grayscale gradient values ​​in the horizontal and vertical directions are respectively obtained by performing weighted differential convolution calculation on the grayscale pixel values ​​in the neighborhood of each pixel), and summing the square of the grayscale gradient value of each pixel in the horizontal direction and the square of the grayscale gradient value in the vertical direction, and performing square root processing on the summation result to obtain the gradient intensity value of each pixel, and performing matching analysis based on the two-dimensional coordinates of each pixel and the two-dimensional coordinates of each edge pixel to obtain the gradient intensity value of each edge pixel, and performing standard deviation processing to obtain the wound boundary fluctuation abnormality index of the patient to be treated;

[0037] The brightness value of each pixel in the wound image of the patient to be treated is extracted and a corresponding brightness image is constructed. The Sobel operator is used to calculate the brightness gradient map in the horizontal and vertical directions, and the square root formula is used to calculate the brightness gradient amplitude of each pixel to construct the brightness gradient map. Then, a brightness gradient threshold is set (for example, greater than 1.5 times the local average gradient value). The brightness gradient map is binarized, and all pixels above the threshold are marked as high-reflectivity pixels. A binary highlight mask map is generated. Then, a connected domain analysis operation is performed to identify all continuous highlight regions. The number of pixels (defined as the area) and the outline boundary length (calculated using the outline extraction algorithm) of each highlight region are counted. For each highlight region, its fragmentation value is calculated (i.e., outline boundary length / number of pixels), which is defined as the ratio of boundary length to area. The average fragmentation and area coefficient of variation (the ratio of the standard deviation of the area of ​​all connected highlight regions to the average) of all highlight regions in the entire image are calculated and weighted to obtain the reflective fragmentation index of the patient's wound to be treated.

[0038] The specific formula for calculating the structural resolution index of the wound surface image of the patient to be treated is as follows: Among them, TqD is the structural resolution index of the wound image of the patient to be treated, PyZ is the spectrum clustering index of the wound image of the patient to be treated, α1 is the clustering coefficient stored in the database, BdC is the abnormal boundary fluctuation index of the wound of the patient to be treated, α2 is the boundary coefficient stored in the database, FgP is the reflective fragmentation index of the wound of the patient to be treated, α3 is the fragmentation coefficient stored in the database, and α4 is the interaction coefficient stored in the database.

[0039] It should be explained that α1, α2, α3, and α4 can be obtained through the following steps: using historical data, combined with the wound image spectral segregation index, wound boundary fluctuation anomaly index, and wound reflective fragmentation index, statistical regression analysis is performed to quantify the specific impact of each factor on the wound image structure resolution index, thereby fitting the initial weight value; secondly, setting different value intervals for the above three variables, while keeping other parameters unchanged, gradually adjusting their values ​​and calculating the variation range of the wound image structure resolution index to test the influence trend of each parameter disturbance on the output result, thereby verifying the stability and applicability of the weight setting on the model output, and correcting the value of the initial weight value accordingly.

[0040] In this implementation, multidimensional image quality assessment indicators such as spectral segregation index, boundary fluctuation anomaly index and reflective fragmentation index are extracted by grayscale processing, frequency domain fast Fourier transform and edge detection of each pixel. These indicators reflect the clarity, edge stability and reflective interference of the wound image from different dimensions, so that the structural characteristics of the image can be comprehensively and accurately characterized. Secondly, combined with historical data, the statistical regression method is used to fit the weight coefficients of each indicator to achieve dynamic adjustment and optimization, thereby improving the adaptability and robustness of the model to different wound image qualities, thereby avoiding the one-sidedness of a single indicator, and integrating multiple image quality factors to make the wound image structure resolution index more objective and physically meaningful. Finally, it provides a reliable quality assessment basis for subsequent image enhancement and wound identification, thereby effectively ensuring the scientificity and accuracy of medication recommendations, and significantly improving diagnostic effects and treatment guidance capabilities.

[0041] Specifically, the image enhancement model is a conditional generative adversarial network, including a generator that generates wound-enhanced image data. The generator includes an input encoding layer, a residual modeling layer, a structural attention enhancement layer, an image decoding and reconstruction layer, and a discriminator that distinguishes the generated wound-enhanced image from the real wound image.

[0042] Among them, the specific steps of obtaining the enhanced image data of the wound surface of the patient to be treated are as follows: in the input coding layer of the generator, the wound surface image data is input into the feature coding module composed of a multi-layer convolutional neural network, and each layer of convolution operation is combined with the ReLU activation function and batch normalization operation to extract the local texture features and edge structure information of the image, and generate a low-dimensional feature map tensor; in the residual modeling layer of the generator, the feature map tensor is input into multiple series-connected residual connection modules, each module is composed of two consecutive convolutional layers and short-circuit connections, which are used to model the structural disturbance areas such as blur, occlusion or uneven illumination in the image, so as to improve the generator's response ability to the location of image defects; in the residual modeling layer of the generator, the feature map tensor is input into multiple series-connected residual connection modules, and .... In the structural attention enhancement layer of the generator, an attention mechanism module is introduced based on the feature map output by the residual modeling module. It is used to guide the network to give higher weights to key areas in the image (such as edge breaks and high-reflection details) by calculating channel attention weights and spatial attention masks, thereby enhancing their expression in subsequent image reconstruction; in the image decoding and reconstruction layer of the generator, the enhanced feature map is input into the image reconstruction module composed of upsampling operations (such as deconvolution or pixel shuffle) and convolution fusion, and gradually restored to the spatial size consistent with the original input image, and the wound-enhanced image data is output, that is, the enhanced pixel value and enhanced two-dimensional coordinates of each enhanced pixel point in the wound image.

[0043] The input coding layer is used to extract the basic texture and edge features of the wound image.

[0044] The residual modeling layer is used to enhance the structural representation of blurred or occluded areas in the image.

[0045] The structure attention enhancement layer is used to highlight the feature responses of key detail areas in the image.

[0046] The image decoding and reconstruction layer is used to restore the image size and output the enhanced image result.

[0047] The pre-training steps of the conditional generative adversarial network are as follows: obtain several groups of wound image samples, including damaged wound images and corresponding clear wound images, and divide them into wound image training set and wound image verification set; initialize the conditional generative adversarial network, including parameter initialization of the weights and biases of the generator and discriminator (that is, use the Kaiming initialization method to initialize the normalized distribution of the weights of each convolutional layer, and initialize all bias terms to zero to ensure that the network has stable gradient propagation ability in the initial state); set the training rounds based on the wound image training set, and perform training loop processing, that is, update the generator and discriminator in each training loop (generator update processing Gradient backpropagation is performed based on the joint optimization objective of image reconstruction loss and adversarial loss, and the discriminator update process is optimized by cross-entropy loss based on the results of real and fake image discrimination); after each training cycle, an evaluation analysis is performed based on the wound image verification set (i.e., based on the image clarity score, the visual quality of the generated image and the corresponding clear wound image is quantitatively evaluated), and the parameters of the conditional generative adversarial network are adjusted based on the evaluation analysis (i.e., the learning rate, weight decay factor or loss function weight ratio are adjusted according to the changing trend of the image clarity score to improve the generalization ability and stability of the network in enhancing image quality) until the image enhancement result output by the conditional generative adversarial network meets the preset image quality standard.

[0048] In this implementation, a conditional generative adversarial network structure is used, combined with the collaborative training of the generator and the discriminator, to achieve high-quality enhancement of wound images. Secondly, the generator extracts local texture and edge features through a multi-layer convolutional neural network. The residual modeling layer effectively repairs problems such as blur, occlusion, and uneven lighting in the image. The structural attention enhancement layer focuses on key areas, such as edge fractures and highly reflective details, thereby significantly improving the detail expression ability of the enhanced image. The image decoding and reconstruction layer restores the spatial size consistent with the original input through upsampling and convolutional fusion, ensuring the naturalness and integrity of the enhanced image. Finally, during the pre-training process, weight initialization and multi-round joint optimization are used to continuously improve the model performance based on image reconstruction loss and adversarial loss. The training parameters are dynamically adjusted based on the clarity score of the validation set to enhance the generalization ability and stability of the model, thereby improving image clarity and structural integrity, reducing blur and noise interference, and significantly improving the accuracy of the wound response index and the scientific nature of medication recommendations, thereby providing more accurate auxiliary support for clinical wound diagnosis and treatment.

[0049] Specifically, the wound enhancement image data is the enhanced pixel value and enhanced two-dimensional coordinates of each enhanced pixel point in the wound image. The specific steps for analyzing the trauma response index of the patient to be treated are as follows: the wound enhancement image data of the patient to be treated is input into the pre-trained wound recognition model for feature extraction and analysis to obtain the wound status feature set of the patient to be treated, including the edge regeneration tension fracture index, heterochromatic cluster offset index, reflection path continuity loss index, and structural halo interference index of each wound area; the wound assessment set of the patient to be treated is comprehensively analyzed to obtain the comprehensive trauma status index of each wound area of ​​the patient to be treated, and the mean is processed to obtain the trauma response index of the patient to be treated.

[0050] The specific formula for calculating the comprehensive index of the wound status of a certain wound area of ​​the patient to be treated is as follows: Among them, CsT is the comprehensive index of the trauma status of a certain wound area of ​​the patient to be treated, BzS is the edge regeneration tension fracture index of a certain wound area of ​​the patient to be treated, β1 is the edge coefficient stored in the database, YsC is the heterochromatic cluster offset index of a certain wound area of ​​the patient to be treated, β2 is the offset coefficient stored in the database, JyC is the reflection path continuity loss index of a certain wound area of ​​the patient to be treated, β3 is the loss coefficient stored in the database, ScG is the structural halo interference index of a certain wound area of ​​the patient to be treated, and β4 is the interference coefficient stored in the database.

[0051] It should be explained that β1, β2, β3, and β4 can be obtained through the following steps: using historical data, combined with the edge regeneration tension fracture index, heterochromatic cluster offset index, reflection path continuity loss index, and structural halo interference index, statistical regression analysis is performed to quantify the specific impact of each factor on the comprehensive index of trauma status, thereby fitting the initial weight value. Secondly, different value intervals of the above four variables are set. While keeping other parameters unchanged, their values ​​are gradually adjusted and the change range of the comprehensive index of trauma status is calculated to test the impact trend of each parameter disturbance on the output result, thereby verifying the stability and applicability of the weight setting on the model output, and correcting the value of the initial weight value accordingly.

[0052] The specific implementation example of calculating the comprehensive index of the wound state of a certain wound area of ​​the patient to be treated is as follows. The following data are available: the edge regeneration tension fracture index, heterochromatic cluster deviation index, reflection path continuity loss index, and structural halo interference index of 5 wound areas (randomly selected) of the patient to be treated, as shown in Table 1 and Figure 3-6 As shown:

[0053] Table 1 Example of the quality assessment index set data for the serial image of the wound area of ​​the patient to be treated with medication

[0054]

[0055] The marginal coefficient β1 stored in the database is approximately: 0.386;

[0056] The offset coefficient β2 stored in the database is approximately: 0.612;

[0057] The missing coefficient β3 stored in the database is approximately: 0.481;

[0058] The interference coefficient β4 stored in the database is approximately: 0.624;

[0059] Substituting the data in Table 1 and the above coefficients into the specific formula for calculating the comprehensive index of the wound status of a certain wound area of ​​the patient to be treated with the drug, the following is obtained:

[0060] The comprehensive index of the wound state of the first wound area of ​​the patient to be treated is ln(1+√(0.461 0.386 +0.326 0.612 +0.513 0.481 +0.318 0.624 ))≈0.943;

[0061] The comprehensive index of the wound status of the second wound area of ​​the patient to be treated = ln(1+√(0.297 0.386 +0.413 0.612 +0.276 0.481 +0.437 0.624 ))≈0.928;

[0062] The comprehensive index of the wound status of the third wound area of ​​the patient to be treated = ln(1+√(0.384 0.386 +0.289 0.612 +0.346 0.481 +0.537 0.624 ))≈0.940;

[0063] The comprehensive index of the wound state of the fourth wound area of ​​the patient to be treated = ln(1+√(0.317 0.386 +0.374 0.612 +0.424 0.481 +0.438 0.624 ))≈0.973;

[0064] The comprehensive index of the wound status of the fifth wound area of ​​the patient to be treated is ln(1+√(0.534 0.386 +0.431 0.612 +0.573 0.481 +0.315 0.624 ))≈0.964.

[0065] In this embodiment, by inputting wound enhanced image data into a pre-trained wound recognition model, multidimensional structural features including edge regeneration tension fracture index, heterochromatic cluster offset index, reflection path continuity loss index and structural halo interference index are systematically extracted, thereby effectively reflecting the multi-faceted pathological conditions of the wound area. Statistical regression analysis methods are used to quantitatively fit the relationship between the above indicators and the comprehensive index of wound status, and the weight of each factor can be scientifically determined to achieve an accurate comprehensive evaluation of complex wound characteristics. Finally, by setting the value range of each variable and dynamically adjusting the parameters, the model is ensured to have good robustness to the disturbance of input features, can adapt to the diversity of different wound image quality and pathological manifestations, thereby reducing the risk of misjudgment and missed judgment, and also promoting the scientificity and credibility of the wound response index, thereby significantly enhancing the refinement and personalization of wound pathology assessment, and improving the targetedness of medication recommendations and efficacy prediction capabilities.

[0066] Specifically, the wound recognition model is an image structure neural network (built on the basis of an image semantic segmentation network and an image structure regression network), including an input layer, a structure coding layer, a region segmentation layer, and a feature output layer. The specific steps for obtaining the wound status feature set of the patient to be treated are as follows: in the input layer of the image structure neural network, the wound enhanced image data of the patient to be treated (the enhanced pixel value and enhanced two-dimensional coordinates of each enhanced pixel point in the wound image) are received and preprocessed; in the structure coding layer of the image structure neural network, the preprocessed wound enhanced image data of the patient to be treated is encoded to obtain the wound structure perception vector and the structure feature map tensor of the patient to be treated; in the region segmentation layer of the image structure neural network, the wound structure perception vector and the structure feature map tensor of the patient to be treated are segmented to obtain the structure of several wound regions of the patient to be treated. feature vector; in the feature output layer of the image structure neural network, the structural feature vector of each wound area of ​​the patient to be treated is extracted and processed to obtain the edge regeneration tension fracture index of each wound area of ​​the patient to be treated (characterizing the tension distribution stability and boundary continuity change amplitude of the edge structure of the wound area), the heterochromatic cluster offset index (characterizing the spatial aggregation degree and offset trend of the abnormal color blocks appearing inside the wound area, to indicate the regional evolution risk of exudate diffusion, necrotic tissue migration or abnormal pigment deposition), the reflection path continuity loss index (characterizing the integrity of the tissue reflection layer structure inside the wound area, to indicate the difficulty of wound repair and deep structure damage), and the structural halo interference index (characterizing the intensity and diffusion range of the light and shadow interference of the deep structure of the wound area, to indicate the degree of recognition obstacle caused by exudate coverage, background reflection and texture blur to structural interpretation).

[0067] The specific steps of the encoding process are as follows: 2D convolution kernels of sizes 3×3, 5×5, and 7×7 are used to process the wound enhancement image in parallel, and the following three types of structural feature maps are extracted respectively: the 3×3 convolution kernel mainly extracts sharp fracture edges and short-scale color difference jump bands in the wound area for the basic features of fracture structure; the 5×5 convolution kernel extracts color patch transition bands and weak edge fusion bands in the mid-scale area for constructing the basis of heterochromatic aggregation distribution; the 7×7 convolution kernel covers a larger receptive field and extracts reflective transition textures and halo disturbance structure edges in the background area to provide background interference response information. The three sets of feature maps each contain 64 channels, which are merged to generate a 192-channel structural feature map tensor and saved.

[0068] Batch Normalization is performed on the 192-channel structural feature map tensor to improve the stability of feature distribution. The number of channels is compressed to 64 through a 1×1 convolutional layer to reduce redundant feature channels. For example, the compression process retains the following features: edge angle mutation map, regional grayscale mean offset map, high-frequency texture amplitude map, low-frequency brightness attenuation trend map, etc. Then, the Squeeze-and-Excitation channel attention mechanism is introduced to the compressed 64-channel feature map. The response mean of each channel is obtained through global average pooling, and then input into a two-layer fully connected network for weighted analysis. This improves the expression strength of the following features: reflective jump response channel (used to determine high-reflection broken areas), heterochromatic boundary distribution channel (used to extract heterochromatic cluster density), and color residual texture channel (used to determine texture damage and reflection offset). The weighted feature map retains the main response direction of structural disturbances and edge abnormal changes in various regions.

[0069] Finally, a Global Average Pooling operation was performed on the above 64-channel weighted feature map, and the responses of each channel in the entire image were averaged and summarized to obtain a one-dimensional vector of length 64, which was used as the wound surface structure perception vector. This vector includes features reflecting the edge rupture intensity and continuity index, features reflecting the distribution density and offset direction of the color mutation area, features reflecting the continuity of the collagen reflection path and the abnormal texture distribution, and features reflecting the frequency and distribution diffusion trend of the deep background halo disturbance.

[0070] The specific steps of the segmentation process are as follows: receiving the structure perception vector (a floating-point vector of length 64, representing the global response strength of each channel of the structure feature map tensor) output by the structure coding layer, and inputting the structure perception vector into the channel response adjustment network; the adjustment network consists of a set of normalization functions and channel weighted mapping functions, specifically including: performing maximum normalization processing on each channel value in the structure perception vector, normalizing all channel values ​​to the range of 0-1, and generating a channel response weight coefficient set; then, multiplying the weight coefficient set with the corresponding channel in the structure feature map tensor one by one to obtain a weighted structure response map tensor with the same number of channels as the original image tensor;

[0071] Next, the weighted structural response map tensor and the initially generated mask image are subjected to position mapping processing. Specifically, based on the binary mask image of each wound area, the spatial position index set of the area in the structural map tensor is extracted. Within each index set, the average response value, maximum response value, and response value variance of all channels corresponding to the area are respectively counted, and a local structural statistical feature triplet of the area is constructed. At the same time, the channel value corresponding to the dominant channel of the current mask area (i.e., the channel with the maximum local response value) is extracted from the structural perception vector. Combined with the position information of the channel in the global channel ranking, the global structural response description vector of the current area (for example, including fields such as channel response intensity value, ranking position, and relative response proportion) is constructed.

[0072] The above-mentioned local structural statistical feature triples are concatenated with the corresponding global structural response description vector according to the channel dimension to form the structural feature vector of the current wound area.

[0073] The specific steps of the extraction process are as follows: For the edge regeneration tension fracture index, the features of edge rupture strength and continuity index are extracted and input into a trained multi-layer perceptron (MLP) scoring subnetwork with two hidden layers and normalized output units; the edge regeneration tension fracture index is output, ranging from 0 to 1, with higher values ​​indicating more severe regional boundary mutation, greater fracture tension, and lower healing ability;

[0074] For the heterochromatic cluster shift index, the distribution density and shift direction features of the color mutation area are extracted and input into a regression analysis structure consisting of a convolutional feature extraction unit and a fully connected regression unit. The convolutional feature extraction unit uses a two-dimensional convolution operation with a convolution kernel size of 3×3 and 16 channels to enhance the spatial shift pattern and edge jump trend in the regional color distribution. The convolution output is connected to the ReLU activation function to introduce a nonlinear response. After the Flatten operation, it is sent to a fully connected regression layer with 8 neurons to numerically model the degree of heterochromatic shift in the region. The regression output is normalized by the Sigmoid function to obtain a heterochromatic cluster shift index ranging from 0 to 1. This index is used to represent the shift trend and spatial distribution discreteness of the heterochromatic blocks in the current region. Higher values ​​indicate more severe heterochromatic cluster shift.

[0075] For the reflection path continuity loss index, the continuity degree of the collagen reflection path and the characteristics of the texture abnormal distribution, the feature sub-vector is input into a fracture perception network, including: a channel consistency evaluation module, which calculates the response direction consistency score between the input channel vectors based on the cosine similarity function, reflecting whether the collagen reflection path is locally continuously extended. If the similarity between channels is low, it means that the reflection path direction has jumped. It also includes a texture fluctuation scoring module, which consists of a set of lightweight convolutional layers and subsequent fully connected layers. It automatically extracts the internal statistical change pattern of the feature for modeling abnormal structures such as local texture fragmentation and reflection signal fracture. The outputs of the two modules are spliced ​​and fused and input into the final scoring layer (consisting of a fully connected network with an output dimension of 1), outputting the reflection path continuity loss index, which is normalized to the interval 0-1;

[0076] For the structural halo interference index, the frequency and distribution diffusion trend characteristics of the deep background halo disturbance are extracted and input into the disturbance scoring network guided by Gaussian blur kernel, including: a disturbance pattern simulation subnetwork, which acts on the feature through a set of fitted Gaussian filter templates to generate three sets of disturbance response maps. This process simulates the blur attenuation trend of the regional features under background disturbances of different intensities, which is used to model the potential impact of background blur interference; it also includes a light and shadow disturbance recognition subnetwork, which consists of three layers of small convolutional layers. The input is the above-mentioned disturbance response map, and the spatial drift sensitive features are enhanced through residual connections. The network jointly models the disturbance kurtosis, diffusion trend, frequency distribution and other patterns, and finally outputs a disturbance risk score, which is finally input into the normalization function to adjust it to a fixed numerical range of 0-1. The final output is the structural halo interference index of the wound area.

[0077] The input layer is used to receive and preprocess wound enhancement image data.

[0078] The structure encoding layer is used to extract the structure feature map tensor and structure perception vector.

[0079] The region segmentation layer is used to divide the image into regions based on the structure perception vector and extract the structural feature vector of each region.

[0080] The feature output layer is used to analyze the regional structural feature vectors and calculate various structural state indices.

[0081] And the pre-training steps of the image structure neural network are as follows:

[0082] A dataset of labeled wound surface structural images was obtained. The dataset was generated by medical experts based on clinical trauma image annotation and contains wound surface image samples taken from multiple angles and under multiple lighting conditions. For each image, structural area masks (including edge regeneration zones, heterochromatic tissue clusters, collagen reflection areas, and background halo areas) and corresponding structural status labels (such as abnormal edge tension, heterochromatic migration trends, reflection faults, and the degree of light and shadow interference) were annotated. The dataset was then divided into a structural training set and a structural validation set.

[0083] Kaiming initialization is used to initialize the parameters of the image structure neural network, and convolutional backbone modules pre-trained on image recognition datasets such as ImageNet or MedSeg can be optionally loaded to enhance the model's convergence efficiency and structural perception capabilities in medical images.

[0084] The model training process is performed on the structural training set. The number of training rounds is set (e.g., 120 rounds), and the following steps are performed in sequence in each training round: In the forward propagation stage, the input enhanced wound image is sequentially passed into each layer of the image structural neural network, and the output is the edge regeneration tension fracture index, heterochromatic cluster offset index, reflection path continuity loss index, and structural halo interference index. In the loss function construction stage, a multi-task loss function is designed for the above four types of structural indices: regression difference loss (L2 regression) for the edge index, offset field prediction loss (L1 loss based on color channel offset) for the heterochromatic index, similarity loss (based on channel cosine similarity) for the reflection path loss index, and regional perturbation recognition loss (cross entropy loss based on fuzzy region classification) for the halo interference index. These four losses are weighted and combined according to preset weights to form the total loss function. In the backpropagation and parameter optimization stage, the gradient of the total loss is backpropagated, and the network parameters are updated using the Adam optimizer. At the same time, a learning rate cosine annealing mechanism, gradient clipping strategy, and batch normalization operation are introduced to improve the network training stability and generalization ability of complex structural scenes.

[0085] After each round of training, the structural validation set is used for evaluation reasoning, and the prediction results of four types of structural indices are output. The performance evaluation values ​​of the corresponding indicators are calculated respectively (such as the mean square error (MSE) of the edge fracture index and the goodness of fit (R)2 The system calculates the following metrics: (1) the offset accuracy and color IoU of the heterochromatic offset index, (2) the cosine similarity score and structural accuracy of the reflection path break index, and (3) the classification F1 score and interference area IoU of the halo interference index. The system also plots the convergence curve of the loss function and the epoch-by-epoch trend of each structural index evaluation metric. If multiple evaluation metrics show no significant improvement over multiple rounds of training, the early stopping mechanism is triggered to terminate training early.

[0086] After the training is completed, save the final converged model parameter file, fully record the weight parameters of each layer, regional segmentation structure, channel fusion strategy and model structure configuration of the multi-branch exponential output network, and export it.

[0087] In this implementation, wound enhancement images are systematically processed in layers to comprehensively extract multi-dimensional structural features of each wound area, thereby achieving deep fusion analysis from low-level texture and edge information to high-level semantic area division. Secondly, multi-scale convolution kernels are used to extract fine-grained and large-scale features in parallel, effectively capturing various structural abnormalities such as wound edge fractures, color shift, texture damage, and halo disturbances, thereby improving the comprehensiveness and precision of feature expression. The introduction of channel attention mechanism and normalization processing enhances the responsiveness of key structural channels and improves the sensitivity and discrimination of the model for key lesion areas. The segmentation layer combines with the structural perception vector to accurately locate the image area, realizing independent feature extraction of different pathological areas of the wound, enhancing the spatial resolution ability of the model, and thus greatly improving the accuracy and robustness of wound identification, thereby providing a solid data foundation and technical guarantee for the subsequent scientific calculation of the wound response index and medication recommendation, thereby promoting the precise and personalized development of intelligent wound management.

[0088] Specifically, the specific steps of taking the preset medication recommendation measures for the patient to be medicated based on the injury perception intensity index are as follows: comparing and analyzing the injury perception intensity index of the patient to be medicated with the preset injury perception intensity index threshold range;

[0089] If the patient's injury perception intensity index is lower than the preset lower limit of the injury perception intensity index threshold, it is marked as low risk and the first medication recommendation is taken. Specifically, the PLC controller calls the basic spray control instruction to control the spray bottle to perform a single short-term spray operation, and the spray volume is controlled within the minimum set value range. At the same time, the voice module broadcasts a prompt that it has been identified as a mild wound and asks to perform daily spraying. The wound image, recognition result and spray response information are uploaded to the user terminal via WiFi;

[0090] If the patient's injury perception intensity index is within the preset injury perception intensity index threshold range, it is marked as medium risk and the second medication recommendation is taken. Specifically, the PLC controller calls the medium-volume spray control instruction to control the spray bottle to perform two elastic spray actions. The interval between the actions is adaptively set by the system to ensure that the wound area is fully covered. At the same time, the voice module prompts the moderate wound and performs enhanced spraying treatment. The wound image and assessment information corresponding to this level of status are stored in the system for the doctor's subsequent reference;

[0091] If the patient's injury perception intensity index is higher than the preset upper limit of the injury perception intensity index threshold range, it will be marked as high risk and the third medication recommendation measure will be taken. Specifically, the PLC controller automatically triggers the continuous spray action, controls the spray bottle to perform multiple rapid downward and rebound processes, and ensures that the high-risk area is adequately sprayed. A voice prompt is issued to indicate severe wounds, and please focus on treating this area. The current image and assessment information will be synchronized to the remote management platform via WiFi to achieve remote assessment and feedback guidance.

[0092] In this implementation plan, accurate hierarchical management of wound treatment plans is achieved through the judgment of grading thresholds based on the injury perception intensity index, which significantly improves the scientificity and personalization of medication. Secondly, based on the injury intensity, the three risk levels of low, medium and high are automatically divided, corresponding to different spray doses and spraying frequencies, effectively avoiding the waste of resources and insufficient efficacy caused by medication. The automated instruction execution of the PLC controller ensures the accuracy and response speed of the spray action, reduces human intervention, and improves the stability and repeatability of the treatment process. At the same time, voice prompts realize real-time status feedback and medication guidance for patients, enhance patients' cooperation and sense of security, and upload wound images and evaluation data to user terminals and remote management platforms via WiFi, supporting doctors' remote monitoring and personalized regulation, promoting efficient use of medical resources and timely response of medical services, thereby realizing closed-loop control from data analysis to intelligent execution, and further improving the automation, precision and remote level of wound care, thereby providing patients with more scientific, convenient and effective treatment plans.

[0093] See also Figure 7The embodiment of the present invention provides a technical solution: a spray bottle medication recommendation system based on image processing, comprising: an image structure resolution module, configured to obtain wound image data of a patient to be medicated and analyze a structural resolution index of the wound image of the patient to be medicated; a resolution judgment module, configured to judge whether the structural resolution index of the wound image of the patient to be medicated is lower than a preset threshold value of the wound image structure resolution index; an image confidence analysis module, configured to input the wound image data of the patient to be medicated into a pre-trained image enhancement model for enhancement analysis if the structural resolution index of the wound image of the patient to be medicated is lower than the preset threshold value of the wound image structure resolution index, thereby obtaining enhanced wound image data of the patient to be medicated, and analyzing a confidence index of the wound image of the patient to be medicated; a wound recognition analysis module, configured to analyze the wound response index of the patient to be medicated based on the pre-trained wound recognition model, and perform a comprehensive analysis with the wound image confidence index to obtain an injury perception intensity index of the patient to be medicated; and a medication feedback module, configured to take preset medication recommendation measures for the patient to be medicated based on the injury perception intensity index.

[0094] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0095] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A spray bottle medication recommendation method based on image processing, characterized in that: The following steps are involved: Acquiring wound surface image data of patients to be treated with medication, and analyzing the structural resolution index of the wound surface images of patients to be treated with medication; Determining whether the wound surface image structure resolution index of the patient to be treated is lower than a preset wound surface image structure resolution index threshold; If the wound image structure resolution index of the patient to be treated is lower than a preset wound image structure resolution index threshold, the wound image data of the patient to be treated is input into a pre-trained image enhancement model for enhancement analysis to obtain enhanced image data of the wound of the patient to be treated, and the wound image confidence index of the patient to be treated is analyzed; Based on the pre-trained wound recognition model, the wound image data of the patient to be treated is enhanced, and the trauma response index of the patient to be treated is analyzed. The wound image confidence index is then comprehensively analyzed to obtain the trauma perception intensity index of the patient to be treated. Based on the perceived injury intensity index, pre-set medication recommendations are taken for patients taking medication.

2. The method for recommending medication in a spray bottle based on image processing according to claim 1, characterized in that: The wound image data specifically includes the pixel value and two-dimensional coordinates of each pixel point in the wound image. The specific steps for analyzing the structural resolution index of the wound image of the patient to be treated are as follows: Based on the wound image data of patients to be treated, the image evaluation set of patients to be treated was analyzed, including the wound image spectrum deviation index, the wound boundary fluctuation abnormality index, and the wound reflective fragmentation index; The image evaluation set of patients to be treated with medication was comprehensively analyzed to obtain the structural resolution index of the wound images of patients to be treated with medication.

3. The method for recommending medication in a spray bottle based on image processing according to claim 2, characterized in that: The specific formula for calculating the structural resolution index of the wound surface image of the patient to be treated is as follows: Among them, TqD is the structural resolution index of the wound image of the patient to be treated, PyZ is the spectrum clustering index of the wound image of the patient to be treated, α1 is the clustering coefficient stored in the database, BdC is the abnormal boundary fluctuation index of the wound of the patient to be treated, α2 is the boundary coefficient stored in the database, FgP is the reflective fragmentation index of the wound of the patient to be treated, α3 is the fragmentation coefficient stored in the database, and α4 is the interaction coefficient stored in the database.

4. The method for recommending medication in a spray bottle based on image processing according to claim 1, characterized in that: The image enhancement model is specifically a conditional generative adversarial network, including a generator that generates wound-enhanced image data. The generator contains an input encoding layer, a residual modeling layer, a structural attention enhancement layer, an image decoding and reconstruction layer, and a discriminator that distinguishes the generated wound-enhanced images from the real wound images.

5. The method for recommending medication in a spray bottle based on image processing according to claim 4, characterized in that: The pre-training steps of the conditional generative adversarial network are as follows: Obtain several sets of wound image samples, including damaged wound images and corresponding clear wound images, and divide them into a wound image training set and a wound image verification set; Initialize the conditional generative adversarial network, including initializing the weights and biases of the generator and discriminator; Based on the wound image training set, set the training rounds and perform training cycle processing, that is, update the generator and discriminator in turn in each training cycle; After each training cycle, an evaluation and analysis is performed based on the wound image verification set, and the parameters of the conditional generative adversarial network are adjusted based on the evaluation and analysis until the image enhancement results output by the conditional generative adversarial network meet the preset image quality standards.

6. The method for recommending medication in a spray bottle based on image processing according to claim 1, characterized in that: The wound surface enhanced image data specifically includes the enhanced pixel value and enhanced two-dimensional coordinates of each enhanced pixel point in the wound surface image. The specific steps for analyzing the wound response index of the patient to be treated are as follows: The enhanced image data of the wound surface of the patient to be treated is input into the pre-trained wound recognition model for feature extraction and analysis to obtain the wound surface state feature set of the patient to be treated, including the edge regeneration tension fracture index, heterochromatic cluster offset index, reflection path continuity loss index, and structural halo interference index of each wound area; A comprehensive analysis is performed on the wound assessment set of the patient to be treated to obtain a comprehensive index of the wound status of each wound area of ​​the patient to be treated, and the mean is processed to obtain the wound response index of the patient to be treated.

7. The method for recommending medication in a spray bottle based on image processing according to claim 6, characterized in that: The wound recognition model is specifically an image structure neural network, including an input layer, a structure encoding layer, a region segmentation layer, and a feature output layer. The specific steps for obtaining the wound status feature set of patients to be treated are as follows: In the input layer of the image structure neural network, the wound surface enhanced image data of the patient to be treated with the drug is received and preprocessed; In the structure coding layer of the image structure neural network, the pre-processed enhanced image data of the wound surface of the patient to be treated is coded to obtain the wound surface structure perception vector and the structure feature map tensor of the patient to be treated; In the regional segmentation layer of the image structure neural network, the wound surface structure perception vector and the structural feature map tensor of the patient to be treated are segmented to obtain the structural feature vectors of several wound surface areas of the patient to be treated; In the feature output layer of the image structure neural network, the structural feature vector of each wound area of ​​the patient to be treated is extracted and processed to obtain the edge regeneration tension fracture index, heterochromatic cluster offset index, reflection path continuity loss index, and structural halo interference index of each wound area of ​​the patient to be treated.

8. The method for recommending medication in a spray bottle based on image processing according to claim 1, characterized in that: The specific formula for calculating the injury perception intensity index of patients to be treated with medication is as follows: SgZ=ln(1+η1*CxY)*exp(-η2*TzX)*[1+tanh(η3*CxY*TzX)]; Among them, SgZ is the injury perception intensity index of the patient to be treated, CxY is the trauma response index of the patient to be treated, η1 is the trauma response coefficient stored in the database, TzX is the wound image confidence index of the patient to be treated, η2 is the confidence coefficient stored in the database, and η3 is the synergy coefficient stored in the database.

9. The method for recommending medication in a spray bottle based on image processing according to claim 1, characterized in that: The specific steps for implementing pre-set medication recommendations for patients taking medication based on the Injury Perception Intensity Index are as follows: Comparing and analyzing the sadness perception intensity index of the patient to be treated with the preset sadness perception intensity index threshold range; If the patient's injury perception intensity index is lower than the preset injury perception intensity index threshold, the patient is marked as low risk and the first medication recommendation is taken; If the patient's injury perception intensity index is within the preset injury perception intensity index threshold range, it is marked as medium risk and the second medication recommendation is taken; If the patient's injury perception intensity index is higher than the upper limit of the preset injury perception intensity index threshold range, the patient is marked as high risk and the third medication recommendation measure is taken.

10. A spray bottle medication recommendation system based on image processing, applying the spray bottle medication recommendation method based on image processing according to any one of claims 1 to 9, characterized in that: include: An image structure resolution module is used to obtain wound surface image data of patients to be treated with medication and analyze the wound surface image structure resolution index of patients to be treated with medication; A discrimination and judgment module is used to judge whether the wound surface image structure resolution index of the patient to be treated is lower than a preset wound surface image structure resolution index threshold; An image confidence analysis module is configured to input the wound image data of the patient to be treated into a pre-trained image enhancement model for enhancement analysis if the wound image structure resolution index of the patient to be treated is lower than a preset wound image structure resolution index threshold, thereby obtaining enhanced wound image data of the patient to be treated, and analyzing the wound image confidence index of the patient to be treated; The wound surface recognition and analysis module is used to enhance the wound surface image data of the patient to be treated based on the pre-trained wound surface recognition model, analyze the trauma response index of the patient to be treated, and conduct a comprehensive analysis with the wound surface image confidence index to obtain the trauma perception intensity index of the patient to be treated; The medication feedback module is used to take preset medication recommendation measures for patients taking medication based on the injury perception intensity index.

Images