Color signal identification system based on colorimetric dressing

Abstract

The invention discloses a color signal identification system based on a colorimetric dressing, and the system comprises a composite dressing with an asymmetric structure, and the composite dressing comprises a functional layer which can generate visual color change according to the pH value of a contact liquid; the mobile terminal is used for collecting an image of the composite dressing attached to a target surface; the image processing module is deployed on the mobile terminal and is used for calling the trained image recognition model and carrying out automatic positioning and color feature extraction on a dressing area in the image; and the result analysis module is used for calculating and analyzing the extracted color features and outputting a corresponding classification result based on a mapping relation with a preset pH range. According to the invention, the Janus colorimetric dressing with one-way liquid guide capability and the special mobile terminal anti-interference image recognition model are deeply integrated, so that long-term stability of a signal source, high fidelity of signal acquisition and accuracy and reliability of signal analysis are cooperatively realized in a portable system.

Description

Technical Field

[0001] This invention belongs to the technical field of interdisciplinary applications of smart materials and computer vision, and particularly relates to a color signal recognition system based on colorimetric dressings. Background Technology

[0002] In fields such as industrial inspection, environmental monitoring, and biosensing, smart materials that produce visible color changes in response to specific chemical or physical stimuli (hereinafter referred to as "colorimetric materials"), combined with portable image analysis technology, constitute an important type of real-time detection solution. The core objective of this solution is to indirectly reflect specific parameters of the target environment (such as pH value, temperature, specific ion concentration, etc.) through changes in material color, and to use mobile terminals (such as smartphones) to complete convenient signal acquisition and preliminary analysis.

[0003] In scenarios requiring visual monitoring of pH levels in liquid environments, existing technologies typically employ a two-part system: first, a colorimetric material (often in the form of hydrogels or fibrous membranes) with pH-responsive color-changing properties, whose color changes with the pH of the liquid it comes into contact with; and second, an external color analysis device or algorithm used to collect and analyze the material's color to calculate the pH value. A typical application concept involves placing this material in the liquid environment to be tested, capturing an image using a smartphone camera, then analyzing the color information (such as RGB values) in the image using an algorithm, matching it against a preset pH-color reference relationship, and finally outputting an assessment of the pH value.

[0004] However, in practical applications, especially in complex non-laboratory environments, such existing technologies face systemic technical bottlenecks, resulting in insufficient reliability, stability, and ease of use of the monitoring results. These bottlenecks manifest in the following three aspects:

[0005] 1. Insufficient signal source stability and fidelity: Existing colorimetric materials mostly employ methods such as physically blending or simply loading small-molecule pH indicators (e.g., bromocresol green, methyl red) into a flexible matrix. This approach has a fundamental flaw: the binding force between the indicator and the matrix is ​​weak. Under continuous or high-flow-rate liquid immersion and scouring, the indicator is prone to lateral diffusion and longitudinal loss. This directly leads to blurred color change areas, signal intensity decay over time and with repeated use, and even makes it impossible to distinguish whether the color change is due to pH changes or dye loss, severely impacting the reliability of long-term, dynamic monitoring. Furthermore, most materials have a homogeneous hydrophilic structure and cannot actively manage liquids. When excessive liquid soaks the entire material, the color response occurs within the material's three-dimensional volume, rather than a clear, stable two-dimensional surface, resulting in uneven color and unclear response boundaries, introducing inherent noise into image recognition.

[0006] 2. Weak anti-interference capability during signal acquisition: Existing systems heavily rely on general-purpose devices such as smartphones for image acquisition. However, the color information of the original images acquired by general-purpose cameras is easily distorted under interference from complex ambient light, varying shooting angles, and shadows caused by wrinkles or bends on material surfaces. General-purpose color recognition algorithms (such as simple RGB thresholding) are not optimized for the specific scenario of "colorimetric material recognition." They lack effective mechanisms to remove the aforementioned environmental noise and cannot correct color gradients caused by the deformation of the material itself, resulting in large fluctuations and poor repeatability in the analysis results.

[0007] 3. Low system integration, making it difficult to balance portability and accuracy: In pursuit of analytical accuracy, existing technologies sometimes rely on dedicated spectrometers and other equipment, but this completely sacrifices the portability and cost advantages of the solution, making it difficult to deploy in scenarios requiring on-site, real-time detection. If mobile terminals are insisted upon, the aforementioned material limitations and algorithmic shortcomings make it difficult to achieve high-precision, interference-resistant analysis while maintaining portability. Existing solutions often optimize materials or algorithms in isolation, lacking an end-to-end systematic design from "stable signal generation" to "reliable signal acquisition" and then to "accurate signal analysis," leading to the accumulation of errors at each stage and limiting overall performance.

[0008] In summary, the closest existing technologies have inherent defects in terms of material systems, structural design, and system integration, making it difficult for them to provide a portable visual monitoring solution that is stable and reliable in real-world complex application scenarios, has strong anti-interference capabilities in data acquisition, accurate analysis, and is easy to deploy. Summary of the Invention

[0009] To solve the above-mentioned technical problems, the present invention provides a color signal recognition system based on colorimetric dressings, comprising:

[0010] A composite dressing with an asymmetric structure, the composite dressing including a functional layer capable of producing a visual color change according to the pH value of the liquid in contact;

[0011] A mobile terminal for acquiring images of the composite dressing applied to a target surface;

[0012] The image processing module, deployed on the mobile terminal, is used to call the trained image recognition model to automatically locate and extract color features of the dressing area in the image;

[0013] The results analysis module is used to calculate and analyze the extracted color features and output the corresponding classification results based on the mapping relationship with the preset pH range.

[0014] Optionally, the composite dressing includes a porous gel layer and a fiber layer bonded in situ by electrospinning to form a Janus structure with different wettability; the porous gel layer includes metal-organic framework particles loaded with pH-responsive dyes.

[0015] Optionally, the metal-organic framework particles loaded with pH-responsive dyes are phenol red@ZIF-8 particles, which are prepared by adding zinc nitrate hexahydrate solution dropwise to a methanol mixture containing 2-methylimidazole and phenol red, followed by centrifugation and drying.

[0016] Optionally, the image recognition model is built on the YOLOv5s network framework and trained using a dedicated dressing image dataset that includes various lighting, angles, and backgrounds.

[0017] Optionally, the mobile terminal further includes a display module for overlaying the classification results and the positioning frame of the dressing area onto an image or real-time video screen.

[0018] Optionally, the porous gel layer is prepared by freeze-thaw method from polyvinyl alcohol, sodium alginate, polyethylene glycol and phenol red@ZIF-8 particles.

[0019] Optionally, the fiber layer is prepared by electrospinning from a spinning solution comprising polyurethane, curcumin, and silver nitrate.

[0020] On the other hand, the present invention also provides a color signal recognition method based on colorimetric dressings, which, based on the aforementioned system, includes the following steps:

[0021] A composite dressing with an asymmetric structure is provided, the composite dressing including a functional layer capable of producing a visual color change according to the pH value of the liquid in contact;

[0022] Images of the composite dressing applied to the target surface are captured using a mobile terminal;

[0023] The image is input to the image processing module deployed on the mobile terminal, and the trained image recognition model is invoked to automatically locate and extract the color features of the dressing area in the image;

[0024] The extracted color features are calculated and analyzed, and the corresponding classification results are output based on the mapping relationship with the preset pH range.

[0025] On the other hand, the present invention also provides an electronic device including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0026] On the other hand, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0027] Compared with the prior art, the present invention has the following advantages and technical effects:

[0028] To address the signal ambiguity and attenuation issues caused by the easy loss of indicators and disordered liquid management in existing colorimetric materials, this invention employs a synergistic material design combining "metal-organic framework encapsulation" and "Janus asymmetric structure." Specifically, phenol red is loaded into metal-organic frameworks such as ZIF-8 to form composite particles, which are then immobilized in a gel matrix. This design utilizes the microporous structure of the framework to physically confine dye molecules, fundamentally preventing their dissolution and diffusion in the liquid environment. Simultaneously, a hydrophobic fiber layer is constructed in situ on the gel layer using electrospinning technology, forming a Janus structure with a wettability gradient. This structure can actively and rapidly unidirectionally guide the contacting liquid from the response layer, preventing diffusion at the response interface caused by liquid immersion. These methods directly achieve a fundamental innovation in the signal source: the dye retention rate is greatly improved, ensuring high repeatability of the color response throughout the entire service life; at the same time, the directional liquid management confines color changes to a clear, uniform two-dimensional surface, generating a stable visual signal with sharp boundaries suitable for machine recognition, laying a reliable physical foundation for subsequent accurate analysis.

[0029] To address the issue of color information distortion caused by variables such as ambient light and shooting angle during mobile image acquisition, this invention overcomes the limitations of general-purpose algorithms and develops an image recognition model trained on specific dressing image data and deployed in a lightweight manner. The model's construction is deeply integrated with the application scenario: its training dataset was specifically created for the Janus colorimetric dressing described in this invention, covering complex situations with various lighting, angles, and backgrounds; the model structure is optimized to perform real-time and accurate localization of the dressing area, and it incorporates targeted image enhancement and interference correction algorithms. These methods directly overcome the challenge of inaccurate signal extraction in complex environments, enabling the system to effectively remove environmental noise and resist the effects of non-uniform lighting and geometric deformation. This allows for the extraction of true and robust color features strongly correlated with pH values ​​from the original image, significantly improving the objectivity and accuracy of the analysis.

[0030] Ultimately, to overcome the system performance bottlenecks caused by isolated improvements to materials and algorithms, this invention does not address the issue of patching a single component, but rather proposes an end-to-end collaborative design system from "signal generation" to "signal analysis." It deeply integrates the aforementioned stable Janus colorimetric dressing, general-purpose mobile terminal hardware, and a dedicated lightweight recognition model. This system-level solution successfully combines the physicochemical innovation of materials with the information processing of algorithms, resulting in a holistic technical effect: while fully maintaining the portability advantages of smartphones, it simultaneously achieves long-term stability of the signal source, high precision in acquisition and analysis, and strong anti-interference capabilities. This provides a portable, reliable, and user-friendly overall visual monitoring solution, effectively resolving the fundamental contradiction in existing technologies where portability and accuracy are difficult to balance. Attached Figure Description

[0031] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0032] Figure 1 The images show scanning electron microscope (SEM) images, transmission electron microscope (TEM) images, and particle size distribution diagrams of PR@ZIF-8 prepared according to embodiments of the present invention.

[0033] Figure 2 The following is a Fourier Transform Infrared (FTIR) spectrum of PR@ZIF-8 according to an embodiment of the present invention;

[0034] Figure 3 The images show the SEM image and elemental distribution diagram of the PVA-SA-PEG-PR@ZIF-8 gel in this embodiment of the invention.

[0035] Figure 4 Color response diagrams, colorimetric variation diagrams, and Euclidean distance diagrams of PVA-SA-PEG-PR@ZIF-8 gels containing different concentrations of PR@ZIF-8 prepared according to embodiments of the present invention under different pH conditions;

[0036] Figure 5 The pH-dependent color change and quantitative characterization diagram of the PVA-SA-PEG-PR@ZIF-8 gel in this embodiment of the invention;

[0037] Figure 6 This is a graph showing the reversible color change performance of the PVA-SA-PEG-PR@ZIF-8 gel prepared in an embodiment of the present invention under pH cycling.

[0038] Figure 7The diagram shows the experimental results of the liquid permeation efficiency of the HZ-FCA dressing prepared according to an embodiment of the present invention, as well as the corresponding change in water contact angle.

[0039] Figure 8 This is a diagram showing the unidirectional moisture-wicking effect of the HZ-FCA dressing prepared according to an embodiment of the present invention.

[0040] Figure 9 This is a schematic diagram of the YOLO-CDM network structure according to an embodiment of the present invention;

[0041] Figure 10 This is a schematic diagram of a smartphone deployment framework according to an embodiment of the present invention. Detailed Implementation

[0042] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0043] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0044] Example 1

[0045] This embodiment provides a color signal recognition system based on colorimetric dressings. Its main objectives include: providing a high-quality, high-stability visual monitoring method; optimizing the immobilization of pH indicators in the substrate to improve color change reliability; optimizing the dressing structure to form a clear, uniform, and stable visual signal; providing an anti-interference, high-precision, portable intelligent analysis method by developing and deploying a dedicated lightweight deep learning model (YOLO-CDM (Colorimetric Dressing Monitoring)) as an image recognition model, deployed to an application named WoundAI on a smartphone terminal, to achieve automatic positioning of dressing images, environmental noise correction, and intelligent state classification; and achieving end-to-end system optimization of "signal generation-acquisition-analysis" to improve the performance of the integrated monitoring system.

[0046] The system specifically includes:

[0047] A composite dressing with an asymmetric structure, the composite dressing including a functional layer capable of producing a visual color change according to the pH value of the liquid in contact;

[0048] A mobile terminal for acquiring images of the composite dressing applied to a target surface;

[0049] The image processing module, deployed on the mobile terminal, is used to call the trained image recognition model to automatically locate and extract color features of the dressing area in the image;

[0050] The results analysis module is used to calculate and analyze the extracted color features and output the corresponding classification results based on the mapping relationship with the preset pH range.

[0051] The composite dressing is feasible in that it comprises a porous gel layer and a fiber layer bonded in situ by electrospinning to form a Janus structure with different wettability; the porous gel layer includes metal-organic framework particles loaded with pH-responsive dyes.

[0052] Furthermore, the metal-organic framework particles loaded with pH-responsive dyes are phenol red@ZIF-8 particles, which are prepared by adding zinc nitrate hexahydrate solution dropwise to a methanol mixture containing 2-methylimidazole and phenol red, followed by centrifugation and drying.

[0053] Furthermore, the porous gel layer is prepared by freeze-thaw method using polyvinyl alcohol, sodium alginate, polyethylene glycol and the phenol red@ZIF-8 particles.

[0054] Furthermore, the fiber layer is prepared by electrospinning from a spinning solution comprising polyurethane, curcumin, and silver nitrate.

[0055] As one specific implementation method, the process of preparing the spinning solution includes:

[0056] Preparation of PU / Cur / Ag solution: Polyurethane (PU), curcumin (Cur), and silver nitrate (AgNO3) were mixed in a ratio of 100:15:5. The solvent system was a mixture of acetone and N,N-dimethylformamide (mass ratio of 4:6). The resulting mixture was placed in a brown reagent bottle and stirred at a constant speed on a magnetic stirrer overnight at room temperature to obtain a homogeneous solution.

[0057] The preparation process of PVA-SA-PEG-PR@ZIF-8 gel includes:

[0058] (1) Preparation of PR@ZIF-8:

[0059] 2-Methylimidazole was dissolved in methanol and then a certain concentration of PR solution was added; Zn(NO3)2·6H2O was dissolved in water and added dropwise to the above 2-methylimidazole system under stirring. After reacting for 30 min, the mixture was centrifuged, the precipitate was collected and dried at 60 °C to obtain PR@ZIF-8 particles.

[0060] (2) Preparation of PVA-SA-PEG-PR@ZIF-8 gel:

[0061] PR@ZIF-8 was uniformly dispersed in a mixed solution of deionized water and ethylene glycol (volume ratio 3:1); polyvinyl alcohol (PVA), sodium alginate (SA), and polyvinyl alcohol (PEG) were weighed at a mass ratio of 100:16:75 and dissolved in the PR@ZIF-8 solution; the mixture was stirred at 95℃ for 2 hours to obtain a uniform prepolymer solution; then it was poured into a mold, placed in a -20℃ freezer for 12 hours, and then thawed at room temperature for 2 hours to obtain PVA-SA-PEG-PR@ZIF-8 gel;

[0062] Furthermore, the specific preparation method of PR@ZIF-8 is as follows:

[0063] Dissolve 330 mg of 2-MI in 5 mL of methanol; slowly add 5 mL of 0.06% (v / v) PR solution while stirring, ensuring thorough mixing. Separately, dissolve 150 mg of Zn(NO3)2·6H2O in 5 mL of deionized water. Add the Zn(NO3)2·6H2O solution dropwise to the 2-MI / PR mixture while magnetically stirring, continuing stirring for 30 min. Then, centrifuge the reaction mixture three times (15 min each time), discard the supernatant, collect the precipitate, and dry it at 60 °C to obtain PR@ZIF-8 particles.

[0064] Furthermore, the amount of PR@ZIF-8 added is (0-0.1) wt% of the total mass.

[0065] The preparation process of Janus composite dressing with one-way fluid guiding and pH responsive functions includes:

[0066] The PVA-SA-PEG-PR@ZIF-8 gel was fixed on the receiving device, and the electrospinning device was turned on. Spinning parameters were as follows: injection pump speed 0.06 mm / min, needle type 12G, needle-to-receiving device distance 15 cm, positive voltage 15 kV, negative voltage 2 kV. Spinning was performed for a certain period of time according to these parameters. After spinning, the composite dressing was transferred to a vacuum drying oven to remove residual organic solvents from the fiber membrane, yielding the Janus composite dressing PZ-FCA with unidirectional liquid conduction and pH response functions. The fiber layer was formed directly on the gel surface through in-situ electrospinning, resulting in a tight interfacial bond and a stable Janus structure with unidirectional liquid conduction capability.

[0067] Furthermore, the preparation process of the spinning solution includes:

[0068] PU, Cur, and AgNO3 were mixed according to their mass percentages in an AC / DMF solvent system (volume ratio 4:6). The mixture was then stirred at a constant speed overnight on a magnetic stirrer. The mass ratio of PU, Cur, and AgNO3 was 100:15:5; the mass ratio of PU, Cur, and AgNO3 to solvent was 10:1.5:0.5:100.

[0069] Furthermore, the spinning time is (30-240) min.

[0070] The image recognition model is feasible because it is built on the YOLOv5s network framework and trained using a dedicated dressing image dataset that includes various lighting conditions, angles, and backgrounds.

[0071] Alternatively, the mobile terminal may also include a display module for overlaying the classification results and the positioning frame of the dressing area onto an image or real-time video screen.

[0072] As a specific implementation method, a YOLO-CDM (Colorimetric Dressing Monitoring) model is deployed and run on a smartphone to achieve online recognition of classification results. The prepared Janus composite dressing PZ-FCA with pH response function is attached to the target surface. Images or real-time videos of the dressing are captured via a smartphone, and the pre-integrated YOLO-CDM model is invoked to locate the target area. The YOLO-CDM model is built on the YOLOv5s network framework, including a deep learning inference structure composed of a convolutional backbone network, a feature pyramid fusion module, and a multi-scale detection head. It can automatically locate the dressing area in the input image and intelligently classify the results based on the color change features of the dressing surface. During the detection process, the model: extracts color and texture features at different scales through the backbone network; achieves synergistic enhancement of shallow color features and deep semantic information through the feature fusion structure; and performs multi-scale prediction of the color response of the dressing area in the detection head, outputting the corresponding state classification results. Based on the output of the YOLO-CDM model, the WoundAI smartphone application automatically generates visual detection results, including the dressing target bounding box, color status labels, classification level indicators, and related prompts, which are displayed in real time on the interface. This step enables rapid classification of the target status and provides user-friendly feedback.

[0073] Furthermore, the YOLO-CDM model includes:

[0074] (1) Constructing a training dataset: Acquire several images of the dressing with different shooting backgrounds, different lighting conditions, and multiple angles, and label the target areas and corresponding classification levels in them using LabelImg to construct a training dataset for training.

[0075] (2) Model Optimization and Training: The YOLOv5s network framework was used as the basis to construct the YOLO-CDM model. The aforementioned dedicated dataset was input into the model for training. During the training process, data augmentation strategies, including Mosaic augmentation and random affine transformation, were used to improve the robustness of the model, and a cosine annealing learning rate adjustment strategy was applied to dynamically optimize the training process. Through this training, the model was able to accurately locate the dressing from complex backgrounds and obtain classification results based on its color features.

[0076] (3) Model lightweighting and format conversion: After the model training converges, the final model weight file is converted into the TFLite lightweight format suitable for mobile deployment, in preparation for integration into smartphone applications.

[0077] (4) Mobile Integration and Deployment: The converted lightweight model is integrated into the WoundAI application on a smartphone. This step achieves deep integration between the YOLO-CDM model and the mobile application, enabling the application to call the phone's camera to capture images in real time and use the local model to complete the real-time detection and classification level recognition of dressings, forming a complete closed loop of monitoring functions.

[0078] Compared with the prior art, this embodiment has the following advantages:

[0079] This provides a reliable and superior material foundation for monitoring technology, ensuring clear and stable signals:

[0080] The accuracy of intelligent analysis is highly dependent on the signal quality generated by the front-end dressing. Improvements in this embodiment constitute a key support for the system: excellent management and signal fidelity. The Janus hydrophilic / hydrophobic asymmetric structure, constructed through in-situ electrospinning, generates unidirectional liquid guidance driven by a wettability gradient. This structure ensures that the pH-responsive layer forms a sharply defined, uniformly colored indicator area, greatly improving the accuracy and efficiency of image recognition.

[0081] Enhanced indicator stability and long-term monitoring effectiveness are achieved by employing a metal-organic framework to encapsulate natural pH indicators, effectively preventing dye photodegradation and washout. This ensures highly repeatable and reliable color-changing responses of the dressing throughout its service life, providing a stable signal source for long-term dynamic monitoring.

[0082] High-precision intelligent monitoring and classification:

[0083] This embodiment integrates specific colorimetric dressings with a mobile deep learning model into a complete monitoring system. Its core features include: objective and accurate analysis; a dedicated YOLO-CDM model integrated into a smartphone automatically locates the dressing area and effectively eliminates interference from ambient lighting and shooting angles, directly extracting color and texture features strongly correlated with the target and pH state. This completely overcomes the errors of subjective human judgment, providing repeatable quantitative standards for monitoring results and significantly improving the accuracy and consistency of judgments. Monitoring is portable and real-time; leveraging the smartphone platform, image acquisition and analysis can be performed anytime, anywhere within seconds. Users can quickly obtain visualized results (including target bounding boxes, status labels, and classification levels), driving a shift in monitoring methods from subjective experience to intelligent decision-making, greatly improving the timeliness and convenience of monitoring.

[0084] The preparation process is controllable, the materials are widely available, and it is suitable for large-scale production.

[0085] The preparation process in this embodiment is clear and the process parameters are controllable. It does not rely on high-end equipment and is suitable for industrial scale-up. The raw materials are widely available and the costs are controllable, meeting the requirements of green production and providing technical conditions for the industrialization and promotion of the product.

[0086] Example 2

[0087] This embodiment provides a color signal recognition system based on colorimetric dressings. The system specifically includes: a composite dressing with an asymmetric structure, the preparation process of which includes:

[0088] Preparation of PVA-SA-PEG-PR@ZIF-8 gel:

[0089] Preparation of PR@ZIF-8: 330 mg of 2-MI was dissolved in 5 mL of methanol to obtain a clear solution. 5 mL of 0.06% (v / v) PR solution was slowly added under stirring, ensuring thorough mixing. Separately, 150 mg of Zn(NO3)2·6H2O was dissolved in 5 mL of deionized water to form a homogeneous, colorless solution. The Zn(NO3)2·6H2O solution was added dropwise to the 2-MI / PR mixture under magnetic stirring, and stirring was continued for 30 min. Subsequently, the reaction mixture was centrifuged three times (15 min each time), the supernatant was discarded, the precipitate was collected and dried at 60 °C to obtain PR@ZIF-8 particles.

[0090] Preparation of PVA-SA-PEG-PR@ZIF-8 gel: PR@ZIF-8 in wt% (0, 0.02, 0.04, 0.06, 0.08, 0.1 wt%) was dispersed in a solvent (deionized water / ethylene glycol volume ratio 3:1). 1 g PVA, 0.16 g SA, and 0, 2.5, 5, 7.5, 10 wt% PEG were added to the system. The mixture was stirred at 95 °C for 2 h. The prepolymer solution was poured into a mold and frozen at -20 °C for 12 h, followed by natural thawing at room temperature for 2 h to obtain a PVA-SA-PEG-PR@ZIF-8 gel with a porous structure.

[0091] Preparation of spinning solution:

[0092] Polyurethane (PU), curcumin (Cur), and silver nitrate (AgNO3) were mixed according to their mass percentages in an AC / DMF solvent system (volume ratio 4:6). The mixture was stirred overnight on a magnetic stirrer at a constant speed. The mass ratio of PU, Cur, and AgNO3 was 100:15:5; the mass ratio of PU, Cur, and AgNO3 to solvent was 10:1.5:0.5:100.

[0093] Janus dressing preparation:

[0094] A porous PVA-SA-PEG-PR@ZIF-8 gel was prepared by freeze-drying 0.06 wt% PR@ZIF-8, 1 g PVA, 0.16 g SA, and 7.5 wt% PEG. This gel was fixed onto a receiving device and electrospun using a PU / Cur / Ag solution. Spinning parameters were: injection pump speed 0.06 mm / min, needle type 12G, needle-to-receiver distance 15 cm, positive voltage 15 kV, and negative voltage 2 kV. After spinning, the mixture was transferred to a vacuum drying oven to remove residual organic solvents, yielding the Janus dressing.

[0095] Furthermore, smartphone-assisted monitoring systems based on the YOLO-CDM model include:

[0096] (1) Data collection and annotation:

[0097] Using the prepared Janus dressing, different pH environments were simulated, and images of the dressing were captured using a smartphone under various real-world scenarios (different light intensities, shooting angles, and skin backgrounds). Crucially, all training data originated from the dressing used in this embodiment. Subsequently, annotation tools (such as LabelImg) were used to create bounding boxes for the dressing areas in the images, and the annotations were categorized according to the simulated pH environment to form training and testing datasets.

[0098] (2) Model training and optimization:

[0099] The YOLO-CDM model is built upon the YOLOv5s network framework. The internal processing flow is as follows: The backbone network uses a CSPDarknet structure. Its function is to perform layer-by-layer convolution and downsampling on the input dressing image, extracting multi-scale feature maps ranging from low-level edge and color features to high-level semantic features. The neck network uses a PANet structure. Its function is to perform feature fusion, combining the deep semantic features extracted by the backbone network with shallow high-resolution color and texture features through upsampling and downsampling operations, thereby enhancing the model's ability to distinguish subtle changes in dressing color. The detection head contains prediction layers at multiple scales. Its function is to predict on the fused feature map, outputting the bounding box of the dressing target, its category, and the confidence score of the target's presence within the box. The training objective is to enable the model to accurately and quickly locate the dressing region in complex backgrounds, resist interference from changes in ambient light and shooting angles, and reliably obtain classification results based solely on the color and texture features of the dressing region.

[0100] (3) Model export and mobile deployment:

[0101] After model training is complete, it is lightweighted and deployed to form a usable monitoring system. The trained YOLO-CDM model is optimized and converted to TFLite format to adapt to mobile computing resources. The packaged model is then integrated into the WoundAI smartphone application. This step integrates the YOLO model with the mobile app, making the phone's camera a signal collector and the local model a real-time analysis engine.

[0102] (4) System workflow verification:

[0103] After applying the PZ-FCA dressing to the target area, the user opens the WoundAI app and uses their phone's camera to photograph the dressing. The app automatically invokes the built-in YOLO-CDM model to analyze the captured image in real time. The model performs two core tasks: automatic localization, outlining the dressing's location in the video stream in real time; and state classification, analyzing the color features within the outline and immediately outputting the classification result. The app interface overlays the analysis results onto the real-time screen, including the dressing location outline, clear classification level text labels, and possible brief prompts, completing the entire monitoring process from image acquisition to result output.

[0104] Figure 1 A in the image is a SEM image of PR@ZIF-8, which shows that the particles have a polyhedral morphology, a relatively smooth surface, and a uniform particle size distribution. Figure 1 B in the image is a TEM image, which further confirms the typical metal-organic framework structure of PR@ZIF-8; Figure 1 The C value in the figure shows that the particle size of PR@ZIF-8 is about 161 nm, indicating that the preparation process is stable and the resulting particles are uniformly distributed, which is beneficial to their uniform dispersion in the gel system.

[0105] The FTIR spectrum of PR@ZIF-8 is as follows: Figure 2 As shown, typical characteristic absorption peaks can be observed in ZIF-8, including a CH stretching vibration peak at 2925 cm⁻¹, a CN stretching vibration peak at 1145 cm⁻¹, and a Zn-N coordination vibration peak at 421 cm⁻¹. For the supported phenol red, its characteristic peaks are mainly the aromatic ring C=C stretching vibration peak at 1595 cm⁻¹ and the S=O stretching vibration peak at 1040 cm⁻¹.

[0106] The microstructure and elemental distribution of PVA-SA-PEG-PR@ZIF-8 gel are as follows: Figure 3 As shown, Figure 3 The SEM images show that the gel exhibits a three-dimensional porous network structure with uniform pore distribution. The elemental distribution map (EDS Mapping) shows that Zn, C, O, S and other elements are uniformly dispersed, indicating that PR@ZIF-8 can be stably embedded in the gel system without obvious aggregation.

[0107] The pH response and color change of PVA-SA-PEG-PR@ZIF-8 gel are as follows: Figure 4 As shown, Figure 4 Image A shows optical images of PR@ZIF-8 gels at different concentrations under pH 4–9 conditions. As pH increases, the color gradually changes from yellow to red to purplish-red, which can intuitively reflect the changes of the detection target with the microenvironment. Figure 4 In the figure, B–F represent the colorimetric values ​​corresponding to different concentrations of PR@ZIF-8 gel. Figure 4 In this context, G represents the Euclidean distance variation; the more significant the chromaticity difference, the more sensitive the color response.

[0108] pH response characteristics and quantitative analysis of PVA-SA-PEG-PR@ZIF-8 gel, as follows: Figure 5 As shown, Figure 5 Figure A shows the RGB channel value changes corresponding to the surface color of the gel after immersion in pH 4–9 buffer solutions, indicating that the gel has a clear pH visual response capability, and the response covers the possible physiological pH range. Figure 5 Figure B shows the fitted curve based on the G channel signal intensity and pH value. The most sensitive and stable G channel data was selected for fitting, yielding a good functional relationship. This quantitative relationship confirms that indirect and objective pH measurement can be achieved through simple image analysis (such as extracting the G value), providing a reliable quantitative basis for subsequent intelligent classification using smartphone images.

[0109] The reversibility verification results of PVA-SA-PEG-PR@ZIF-8 gel are as follows: Figure 6 As shown, the reversible response performance of the material was evaluated by alternately placing the gel in buffer environments with different pH values ​​and measuring its RGB values ​​after reaching color equilibrium. The results show that, during multiple cycles, the gel's color signal can stably and reversibly switch between high and low pH states with minimal signal intensity attenuation. This experiment demonstrates that the gel, as an indicator material, possesses good reversibility and long-term stability, ensuring the reliability and repeatability of its signal response during dynamic monitoring, overcoming the shortcomings of traditional indicators such as easy inactivation and irreversible signal attenuation.

[0110] Liquid permeability and unidirectional moisture transfer results, such as Figure 7 As shown, Figure 7 Figure A shows the experimental results of the liquid penetration efficiency of HZ-FCA dressing. Increasing the spinning time will reduce the penetration efficiency. Figure 7 B in the figure represents the change in contact angle during the seepage process, which further verifies this point.

[0111] Figure 8 The results of the unidirectional fluid transfer experiment of HZ-FCA dressing were shown. Water droplets added from the hydrophobic layer could easily penetrate into the gel layer, while water droplets added from the hydrophilic gel side could not pass through in the reverse direction, proving that the dressing has obvious unidirectional fluid transfer characteristics.

[0112] Figure 9 This is a schematic diagram of the YOLO-CDM network structure;

[0113] Figure 10 This paper describes the key steps in integrating a trained YOLO-CDM model into a smartphone application (WoundAI) and implementing real-time monitoring functionality. The steps include model conversion to TFLite format, Android Studio development environment configuration, TFLite model integration into the application, model inference process, application deployment, and feature implementation.

[0114] Furthermore, after launching the installed WoundAI application on their smartphone, users enter the main interface. The main interface provides two core functions: image input and real-time detection. Static image recognition process: After selecting the "Select photos" function on the main interface, users enter the image recognition page. On this page, users can obtain the target image to be recognized in two ways: Take photos: Use the phone's camera to capture the target image in real time and input the captured image into the application's recognition module; Select from local files: Select an existing image file from the phone's local storage and upload the selected image to the application's recognition module.

[0115] After acquiring an image, the application automatically performs subsequent processing steps such as region detection, feature extraction, and recognition analysis. Real-time video stream detection process: After the user selects the "Real-time prediction" function on the main interface, the application will use the phone's camera to acquire a continuous video stream and perform real-time detection and recognition of targets in the video stream. The system performs image processing, detection, and result feedback at the video frame level, achieving real-time analysis of the target's state.

[0116] Example 3

[0117] The present invention also provides a color signal recognition method based on colorimetric dressings, which, based on the aforementioned system, includes the following steps:

[0118] A composite dressing with an asymmetric structure is provided, the composite dressing including a functional layer capable of producing a visual color change according to the pH value of the liquid in contact;

[0119] Images of the composite dressing applied to the target surface are captured using a mobile terminal;

[0120] The image is input to the image processing module deployed on the mobile terminal, and the trained image recognition model is invoked to automatically locate and extract the color features of the dressing area in the image;

[0121] The extracted color features are calculated and analyzed, and the corresponding classification results are output based on the mapping relationship with the preset pH range.

[0122] Example 3

[0123] This embodiment also provides an electronic device, including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0124] Example 4

[0125] This embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0126] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A color signal recognition system based on colorimetric dressings, characterized in that, include: A composite dressing with an asymmetric structure, the composite dressing including a functional layer capable of producing a visual color change according to the pH value of the liquid in contact; A mobile terminal for acquiring images of the composite dressing applied to a target surface; The image processing module, deployed on the mobile terminal, is used to call the trained image recognition model to automatically locate and extract color features of the dressing area in the image; The results analysis module is used to calculate and analyze the extracted color features and output the corresponding classification results based on the mapping relationship with the preset pH range.

2. The system according to claim 1, characterized in that, The composite dressing comprises a porous gel layer and a fiber layer bonded in situ by electrospinning, forming a Janus structure with different wettability; the porous gel layer includes metal-organic framework particles loaded with pH-responsive dyes.

3. The system according to claim 2, characterized in that, The metal-organic framework particles loaded with pH-responsive dyes are phenol red@ZIF-8 particles, which are prepared by adding zinc nitrate hexahydrate solution dropwise to a methanol mixture containing 2-methylimidazole and phenol red, followed by centrifugation and drying.

4. The system according to claim 1, characterized in that, The image recognition model is built on the YOLOv5s network framework and trained using a dedicated dressing image dataset that includes various lighting, angles, and backgrounds.

5. The system according to claim 1, characterized in that, The mobile terminal also includes a display module, which is used to overlay the classification results and the positioning frame of the dressing area onto the image or real-time video screen.

6. The system according to claim 2, characterized in that, The porous gel layer is prepared by freeze-thaw method using polyvinyl alcohol, sodium alginate, polyethylene glycol and phenol red@ZIF-8 particles.

7. The system according to claim 2, characterized in that, The fiber layer is prepared by electrospinning from a spinning solution comprising polyurethane, curcumin, and silver nitrate.

8. A method for color signal recognition based on colorimetric dressings, characterized in that, The system based on any one of claims 1-7 includes the following steps: A composite dressing with an asymmetric structure is provided, the composite dressing including a functional layer capable of producing a visual color change according to the pH value of the liquid in contact; Images of the composite dressing applied to the target surface are captured using a mobile terminal; The image is input to the image processing module deployed on the mobile terminal, and the trained image recognition model is invoked to automatically locate and extract the color features of the dressing area in the image; The extracted color features are calculated and analyzed, and the corresponding classification results are output based on the mapping relationship with the preset pH range.

9. An electronic device comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, characterized in that, When the processor executes the computing program, it implements the method of any one of claims 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 8.

Images