Method, device and storage medium for improving the accuracy of positive and negative identification of test strips

Abstract

The present invention discloses a method, device and storage medium for improving the accuracy of positive and negative identification of test strips. The method comprises: using a multi-spectral camera to collect image data of the C-line and T-line regions of the test strip; performing feature extraction and optimization processing on the collected multi-spectral images; abstracting the detection process into a logic circuit model and constructing a node feature matrix; randomly injecting faults into the model, simulating environmental problems, and generating test vectors; using failure response data to train a graph convolutional neural network model, adding a hole convolution; applying the trained model to analyze real-time data, predict faults and evaluate color changes; if an abnormality is detected, using a color difference matrix and timing analysis to accurately distinguish the red type; calculating the confidence based on the optimized features and high-precision color information, and determining the object frame. By implementing the method of the present invention, the problems of low accuracy, weak anti-interference ability and lack of effective fault simulation in the existing test strip detection technology can be solved.

Description

Technical Field

[0001] The invention relates to a method for identifying the positive and negative properties of a paper strip, and more specifically to a method, a device and a storage medium for improving the accuracy of identifying the positive and negative properties of a test strip. Background Art

[0002] As a fast and simple diagnostic tool, test strip testing is widely used in the medical and health field, such as disease screening, drug abuse detection, environmental monitoring, etc. However, traditional test strip testing methods face many challenges in practical applications, especially the problem of insufficient accuracy in identifying positive and negative results, which directly affects the reliability of test results and the effectiveness of clinical decision-making.

[0003] Traditional test strip testing mainly relies on naked eye observation or simple optical equipment to judge the color changes of the C line (Control Line) and the T line (Test Line) to determine whether the sample is positive. This method has the following limitations: Due to the lack of standardized interpretation standards, there may be large differences in interpretation between different operators. For weakly positive samples, especially when the T line color is very light, it is difficult to accurately identify. External factors such as light and temperature can easily affect the final interpretation results. Traditional methods can usually only provide qualitative results and lack the ability to accurately measure key parameters in the detection process.

[0004] With the development of computer vision technology and machine learning algorithms, some improvement measures have been proposed and applied to test strip detection. For example, image data is acquired through a high-resolution camera and combined with a specific image processing algorithm to improve detection accuracy; different test results are automatically identified and classified using a deep learning model. However, the existing technology still has the following problems: it fails to make full use of multispectral information, resulting in some small but important details being ignored. The lack of effective means to simulate various abnormal situations in the real world limits the learning scope and adaptability of the model. Most systems do not have the function of uploading test results to a cloud database in real time, which is not conducive to subsequent data management and analysis.

[0005] Therefore, it is necessary to design a new method to solve the problems of low accuracy, weak anti-interference ability and lack of effective fault simulation in the existing test strip detection technology. Summary of the invention

[0006] The purpose of the present invention is to overcome the defects of the prior art and provide a method, device and storage medium for improving the accuracy of positive and negative identification of test strips.

[0007] To achieve the above object, the present invention adopts the following technical solution: a method for improving the accuracy of positive and negative identification of test strips, comprising:

[0008] A high-resolution camera is used in combination with three wavelength light sources of red light, green light and near-infrared light to collect multispectral signals of the C line, T line and the area between the test strips to obtain multispectral image data;

[0009] Extracting and optimizing the multispectral image data to obtain an optimized feature vector;

[0010] Based on the optimized feature vector, the test strip detection process is abstracted into a logic circuit model, wherein the key operation steps are used as nodes, the input-output relationship constitutes the edge, and the node feature matrix is ​​constructed. The logic circuit model is used to guide fault injection testing and training data generation;

[0011] Randomly injecting faults at specific node positions in the logic circuit model to simulate problems that may occur in a real environment, generating corresponding test vectors based on the logic circuit model and recording failure response data;

[0012] The graph convolutional neural network model is iteratively trained using the failure response data, and hole convolutions of different sizes are added to multiple network layers to enhance feature extraction capabilities, so that the model can more effectively learn feature representations under different fault modes, and finally a trained graph convolutional neural network model is obtained;

[0013] The graph convolutional neural network model is used to analyze the real-time data of the test strips to be tested, predict possible faults and their probabilities, and evaluate the color changes of the C line and T line of the test strips to preliminarily determine the positive or negative of the test strips;

[0014] If the test result shows positive or there are other abnormal conditions, the color difference matrix and time series analysis method are used to accurately distinguish different types of red in combination with the multispectral image data, and the threshold judgment condition of the reflectance coefficient is automatically adjusted according to the sample characteristics to determine high-precision color change information;

[0015] Based on the optimized feature data and the high-precision color change information, the confidence is calculated, including the coordinates of the object frame, the object confidence, the category of the object frame and the category confidence, the severity level of the object frame and the severity level confidence, and the object frame is filtered and post-processed through techniques such as non-maximum suppression to obtain the target object frame.

[0016] The present invention also provides a device for improving the accuracy of positive / negative identification of a test strip, comprising:

[0017] A data acquisition unit is used to collect multispectral signals of the C line, T line and the area between the test strips using a high-resolution camera combined with three wavelength light sources of red light, green light and near-infrared light to obtain multispectral image data;

[0018] An optimization unit, used for extracting and optimizing features of the multispectral image data to obtain an optimized feature vector;

[0019] A model abstraction unit, used to abstract the test strip detection process into a logic circuit model based on the optimized feature vector, wherein the key operation steps are used as nodes, the input-output relationship constitutes an edge, and a node feature matrix is ​​constructed. The logic circuit model is used to guide fault injection testing and training data generation;

[0020] A fault simulation unit, used for randomly injecting faults into specific node positions in the logic circuit model to simulate problems that may occur in a real environment, generating corresponding test vectors based on the logic circuit model and recording failure response data;

[0021] An iterative training unit, used to iteratively train the graph convolutional neural network model using the failure response data, adding hole convolutions of different sizes in multiple network layers to enhance feature extraction capabilities, so that the model can more effectively learn feature representations under different fault modes, and finally obtain a trained graph convolutional neural network model;

[0022] The fault prediction unit is used to apply the graph convolutional neural network model to analyze the real-time data of the test strip to be tested, predict possible faults and their probabilities, and evaluate the color changes of the C line and T line of the test strip to preliminarily determine the positive or negative of the test strip;

[0023] An analysis unit, for, if the test result shows a positive result or there are other abnormal conditions, using a color difference matrix and a time series analysis method in combination with the multispectral image data to accurately distinguish different types of red, and automatically adjusting a threshold judgment condition of the reflectance coefficient according to sample characteristics to determine high-precision color change information;

[0024] The final prediction unit is used to calculate the confidence based on the optimized feature data and the high-precision color change information, including the coordinates of the object frame, the object confidence, the category of the object frame and the category confidence, the severity level of the object frame and the severity level confidence, and filter and post-process the object frame through techniques such as non-maximum suppression to obtain the target object frame.

[0025] The present invention further provides a computer device, which includes a memory and a processor. The memory stores a computer program, and the processor implements the above method when executing the computer program.

[0026] The present invention also provides a storage medium, characterized in that the storage medium stores a computer program, and the computer program implements the above method when executed by a processor.

[0027] Compared with the prior art, the present invention has the following beneficial effects: the present invention collects multispectral images by combining a high-resolution camera with red, green and near-infrared light sources, thereby enhancing the richness and reliability of image data; uses multispectral image data for feature extraction and optimization, generates optimized feature vectors, and improves the accuracy of model recognition; constructs a fault injection test method based on a logic circuit model, simulates problems in the actual environment by randomly injecting faults, and enhances the robustness of the model; uses a graph convolutional neural network for iterative training, and improves the feature extraction capability of the model under different fault modes through void convolution; accurately distinguishes the type of red through color difference matrix and timing analysis, and automatically adjusts the judgment conditions, thereby further improving the accuracy of color change information and enhancing the model's ability to accurately distinguish test strips.

[0028] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0030] Figure 1 A schematic flow chart of a method for improving the accuracy of positive / negative identification of a test strip provided by an embodiment of the present invention;

[0031] Figure 2 A schematic block diagram of a device for improving the accuracy of positive / negative identification of a test strip provided by an embodiment of the present invention;

[0032] Figure 3 A schematic block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0035] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0036] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0037] See also Figure 1 , Figure 1 A schematic flow chart of a method for improving the accuracy of positive and negative identification of test strips provided in an embodiment of the present invention. The method for improving the accuracy of positive and negative identification of test strips is applied to a server. A multispectral camera is used to collect image data of the C-line and T-line areas of the test strip; feature extraction and optimization processing are performed on the collected multispectral images; the detection process is abstracted into a logic circuit model, and a node feature matrix is ​​constructed; faults are randomly injected into the model to simulate environmental problems and generate test vectors; a graph convolutional neural network model is trained using failure response data and a hole convolution is added; the trained model is used to analyze real-time data, predict faults and evaluate color changes; if an abnormality is detected, a color difference matrix and timing analysis are used to accurately distinguish the type of red; the confidence based on the optimized features and high-precision color information is calculated to determine the object frame; the problems of low accuracy, weak anti-interference ability and lack of effective fault simulation in existing test strip detection technology are solved.

[0038] Figure 1 FIG. 1 is a flow chart of a method for improving the accuracy of positive / negative identification of a test strip provided by an embodiment of the present invention. Figure 1 As shown, the method includes the following steps S110 to S180.

[0039] S110, using a high-resolution camera in combination with three wavelength light sources of red light, green light and near-infrared light to collect multispectral signals of the C line, T line and the area between the test strips to obtain multispectral image data.

[0040] In this embodiment, the multispectral image data refers to an image obtained by collecting multispectral signals of the C line, T line and the area between the test strip using a high-resolution camera in combination with three wavelength light sources of red light, green light and near-infrared light.

[0041] Specifically, high-resolution cameras provide clearer, more detailed image data, ensuring that even subtle color changes or line features are not overlooked. This is especially important for identifying weak positive samples, which may show very faint T-line colors.

[0042] By using three different wavelengths of light sources, red light, green light, and near-infrared light, we can obtain the reaction of the test strip under different spectra. Different wavelengths of light interact with the chemicals on the test strip in different ways, so we can obtain richer information than a single visible spectrum. This helps to distinguish background noise from actual chemical reaction results and improve detection sensitivity.

[0043] Traditional naked-eye observation is easily affected by ambient light conditions, while the use of a fixed light source (such as the specific wavelength light source in this solution) can control the lighting conditions and reduce the risk of misjudgment caused by changes in external light. In addition, near-infrared light is not affected by ambient visible light and can work normally in darker environments, increasing the adaptability and reliability of the system.

[0044] Based on high-resolution images and multispectral data, algorithms can be used to achieve automated, objective interpretation, thereby eliminating subjective differences between different operators. This standardized approach not only improves the consistency of results, but also makes large-scale automated testing possible.

[0045] High-quality multispectral image data provides a solid foundation for subsequent complex image processing steps (such as feature extraction, optimization, etc.) and deep learning model training. Especially for graph convolutional neural networks, multispectral data can help the model better learn the feature representation of various fault modes and further improve the prediction accuracy.

[0046] S120: extracting and optimizing features of the multispectral image data to obtain an optimized feature vector.

[0047] In this embodiment, the optimized feature vector will be used as input for advanced analysis tools such as logic circuit models and graph convolutional neural networks to guide deeper learning and prediction tasks. The optimized feature vector refers to a compact numerical sequence that can effectively represent the state of the C line and T line of the test strip after careful design and optimization. Each value corresponds to a specific feature dimension, reflecting important information about the state of the test strip in that dimension. Such a feature vector not only has good expressive power, but also is easy for computer processing and analysis, thereby providing strong support for achieving high-precision positive and negative identification of test strips.

[0048] In one embodiment, the above-mentioned step S120 may include steps S121 - S124 .

[0049] S121, preprocessing the multispectral image data to obtain a preprocessing result;

[0050] S122, extracting the spatial features, set features and texture features based on color from the preprocessing results to obtain extraction results;

[0051] S123, performing dimension reduction, feature selection, and feature enhancement on the extraction result to obtain an optimization result;

[0052] S124. Construct a feature vector according to the optimization result to obtain an optimized feature vector.

[0053] In this embodiment, first, the acquired multispectral image data needs to go through certain preprocessing steps, including but not limited to operations such as correction, denoising, and normalization, to ensure that images at different wavelengths are consistent and comparable, and to reduce the impact of external factors (such as changes in illumination).

[0054] Since the color changes of the C and T lines of the test strip are the key to judging positive and negative, it is possible to convert from the RGB space to other color spaces (such as HSV, Lab, etc.), which are more conducive to distinguishing color information and extracting color features from them. In addition, the characteristics of the three wavelength light sources of red light, green light and near-infrared light can be used to extract the color features at the corresponding specific wavelengths.

[0055] In addition to color features, geometric features can also be considered, such as the position, length, width, and shape of the lines. For test strips, the position and shape of the C line and T line are very important diagnostic information.

[0056] Texture analysis can help capture subtle structural changes within the line area, which is particularly important for identifying weakly positive samples. Texture features can be extracted using methods such as gray level co-occurrence matrix (GLCM) and local binary pattern (LBP).

[0057] In order to improve computational efficiency and avoid overfitting, techniques such as principal component analysis (PCA), linear discriminant analysis (LDA) or autoencoders can be used to reduce the dimensionality of the extracted high-dimensional features, retaining only the main components that best represent the differences in the data.

[0058] Through correlation analysis, mutual information, recursive feature elimination (RFE) and other methods, the feature subset that is most helpful for the classification task is selected and redundant features are removed.

[0059] The quality of features can be further improved by combining domain knowledge or using deep learning models (such as convolutional neural networks (CNNs)) to automatically learn feature representations. For example, the graph convolutional neural network (GCN) mentioned in this case can automatically learn feature representations that help improve detection accuracy during training.

[0060] Finally, all the above optimized features are combined into a compact and expressive feature vector as the basis for subsequent logic circuit model construction, fault injection testing and training data generation. This feature vector should be able to comprehensively and accurately describe the state of the C line and T line of the test strip and the color change between them, thus providing strong support for the final positive and negative judgment.

[0061] In summary, through comprehensive and detailed feature extraction and optimization of multispectral image data, more accurate and reliable feature vectors can be obtained, thereby significantly improving the accuracy of positive and negative identification of test strips.

[0062] S130. Based on the optimized feature vector, the test strip detection process is abstracted into a logic circuit model, wherein key operation steps are used as nodes, input-output relationships constitute edges, and a node feature matrix is ​​constructed. The logic circuit model is used to guide fault injection testing and training data generation.

[0063] In this embodiment, first, all key operation steps in the test strip detection process need to be identified. These steps may include but are not limited to: sample preparation, test strip insertion, light source irradiation, image acquisition, feature extraction, etc. Each operation step represents a key point in the test strip detection process.

[0064] Then, each of the above key operation steps is defined as a node in the logic circuit model. For example, sample preparation can be an input node; feature vector processing after image acquisition can be regarded as an intermediate node; and the final positive and negative judgment result is the output node.

[0065] Next, determine the input-output relationship between each node, that is, build edges. For example, the result of sample preparation will affect the subsequent test strip insertion operation, which in turn determines the quality of image acquisition, and so on. In this way, a logical connection is established between each step in the entire detection process.

[0066] For each node, a node feature matrix is ​​constructed based on its corresponding optimized feature vector. This matrix contains all relevant feature information of the node, such as color change, geometric shape, texture features, etc. It not only describes the state of the current node, but also reflects the correlation between the previous and next nodes.

[0067] In the constructed logic circuit model, faults are randomly injected at specific node locations to simulate problems that may occur in the real environment. This step aims to explore the behavior of the system under different fault modes, thereby improving the adaptability and robustness of the model to abnormal situations.

[0068] According to the results of fault injection, corresponding test vectors are generated. These test vectors are used to record the failure response data after each fault injection, including but not limited to the degree of color change, position shift, shape change, etc. They will become an important resource for subsequent model training.

[0069] The failure response data obtained above is combined with the data under normal operating conditions to form a rich and diverse training data set. This not only covers common operating scenarios, but also includes various possible abnormal situations, making the trained model more comprehensive and reliable.

[0070] Finally, these carefully designed training datasets are used to iteratively train graph convolutional neural networks (GCNs) or other applicable machine learning models. By continuously adjusting the model parameters, it is ensured that it can accurately learn the feature representations under different failure modes, thereby improving the accuracy of positive and negative identification of test strips.

[0071] In summary, by abstracting the test strip detection process into a logic circuit model and combining fault injection testing with training data generation, the learning scope and adaptability of the model can be effectively improved, ensuring that high detection accuracy is maintained in the face of complex and changing real environments. This method not only helps to improve existing technologies, but also provides new ideas for future research and development.

[0072] S140, randomly injecting faults at specific node positions in the logic circuit model to simulate problems that may occur in a real environment, generating corresponding test vectors based on the logic circuit model and recording failure response data.

[0073] In one embodiment, the above-mentioned step S140 may include steps S141 - S144 .

[0074] S141, determining key nodes and failure modes for the logic circuit model;

[0075] S142, determining a fault injection strategy;

[0076] S143, injecting faults at the key nodes in combination with fault modes according to the fault injection strategy, and generating a test vector including current fault information and its impact range after each fault injection is completed;

[0077] S144. Record failure response data according to the test vector.

[0078] In this embodiment, first, according to the characteristics of the test strip detection process, the key nodes that have a significant impact on the final test result are identified. These nodes may be those places where operational errors, environmental factors or hardware defects are prone to occur.

[0079] For each selected key node, analyze the different types of faults that may be encountered. For example, sample contamination or inaccurate measurement may occur during the sample preparation stage; poor lighting conditions or camera focus issues may occur during image acquisition. Clarifying these fault types will help to perform more accurate fault injection later.

[0080] In order to ensure that fault injection can cover a wide range of possibilities (i.e. randomness) without causing uncontrollable results, a reasonable fault injection strategy needs to be designed. This includes determining factors such as the frequency, intensity, and duration of the fault. At the same time, considering the possible interactions between different faults, it is also necessary to plan how they are combined.

[0081] By using software simulation or hardware modification, fault injection is implemented at the specified node according to the predetermined strategy. For example, the effect of feature extraction can be changed by adjusting algorithm parameters, or external interference (such as temperature change, vibration, etc.) can be directly applied to the physical device to observe its impact on the node output.

[0082] After each fault injection is completed, a test vector containing the current fault information and its impact range is immediately generated. This vector not only records the specific form of the fault (such as type, location, intensity, etc.), but also includes the system state changes caused by it, such as the change trend of certain node output values, the degree of image quality degradation, etc.

[0083] As the test vectors are generated, all failure response data generated by the system during this period are recorded synchronously. This data should be as detailed as possible, including but not limited to:

[0084] The status change trajectory of each node; the difference in final detection results under different failure modes; the time required for the system to recover from a failure; the changes in user interface feedback information, etc.

[0085] The collected failure response data is often messy, so it is necessary to first perform necessary cleaning and standardization to remove outliers, fill in missing values, and convert non-numerical data into a form acceptable to machine learning algorithms.

[0086] Further explore the potential patterns in the failure response data and extract valuable feature variables through feature engineering. This step is crucial to improving the model training effect because it can help the model better understand the internal mechanism of the system when a failure occurs.

[0087] Use the preprocessed and feature-engineered dataset to train the prediction model, evaluate the model performance, and continuously adjust the model structure and parameter settings as needed until satisfactory prediction accuracy is achieved. In addition, the model should be retrained regularly with newly collected data to ensure its long-term effectiveness.

[0088] In summary, by randomly injecting faults into specific node positions in the logic circuit model, generating test vectors based on this and recording failure response data, we can not only gain a deeper understanding of the intrinsic connection between the various steps in the test strip detection process, but also provide strong support for improving existing technologies and improving product quality. This method is conducive to discovering potential design defects and promoting the development and application of new technologies.

[0089] S150. Iteratively train the graph convolutional neural network model using the failure response data, add hole convolutions of different sizes in multiple network layers to enhance feature extraction capabilities, so that the model can more effectively learn feature representations under different fault modes, and finally obtain a trained graph convolutional neural network model.

[0090] In one embodiment, the above-mentioned step S150 may include steps S151 - S153 .

[0091] S151, preprocessing the failure response data to obtain a processing result;

[0092] S152. Construct a multi-layer GCN model. Each layer of the multi-layer GCN model contains standard GCN operations for aggregating neighbor information, and an additional atrous convolution layer for capturing contextual information in a larger range without significantly increasing the number of parameters. Atrous convolution is introduced at different levels of the multi-layer GCN model, and the model is enabled to perceive features of different scales by adjusting the expansion rate. Customized loss functions are used during model training, including object box loss calculation function, object loss function, category loss function, and severity loss function. The severity loss function adds a sample probability penalty term to the cross entropy loss function to increase the model's sensitivity to slight changes.

[0093] S153. Use the processing results to train a multi-layer GCN model to obtain a trained graph convolutional neural network model.

[0094] In this embodiment, a graph structure is constructed according to the characteristics of the test strip positive / negative identification task. Each node represents a detection area on the test strip or its attributes (such as color, position, etc.), and the edge represents the relationship or adjacency between these areas.

[0095] Test strip images in various failure modes are collected and annotated as training samples to ensure that all possible failure types and degrees are covered.

[0096] Perform image preprocessing such as normalization, cropping to a fixed size, and augmentation (such as rotation and flipping) to increase data diversity.

[0097] A multi-layer GCN model is designed, where each layer contains standard GCN operations for aggregating neighbor information and additional hole convolution layers to capture contextual information in a wider range without significantly increasing the number of parameters.

[0098] Atrous convolutions are introduced at different levels of GCN, and the dilation rate is adjusted to allow the model to perceive features of different scales. For example, a smaller dilation rate is used in the shallow layer to obtain local details, while a larger dilation rate is used in the deep layer to expand the receptive field in order to better understand the global information.

[0099] If the task involves locating a specific area on the test strip, it is necessary to define a loss term to measure the difference between the predicted object box and the actual object box, such as using the intersection over union (IoU) loss.

[0100] For the binary classification problem of whether the target object exists, binary cross entropy loss can be used to evaluate the gap between the model output and the actual label.

[0101] When there is multi-class classification, a multi-class cross entropy loss function is applied to ensure that the model correctly assigns the input to the correct fault class.

[0102] For the continuous value estimation of fault severity, a new loss term, severity loss function, is proposed. It is based on cross entropy loss but adds a sample probability penalty term in the calculation process, that is, a heavier penalty is imposed when the predicted result deviates greatly from the true value, thereby increasing the model's sensitivity to subtle changes.

[0103] The GCN model designed above is trained end-to-end iteratively using the prepared failure response dataset. In each iteration, the prediction value is calculated by forward propagation, the total loss is calculated according to the customized loss function, and the weights are updated by backpropagation.

[0104] Experiment with different learning rates, batch sizes, and other hyperparameters to find the best configuration to optimize model performance.

[0105] Regularly evaluate model performance using validation sets, monitor key evaluation indicators (such as accuracy, recall, F1 score, etc.), and adjust strategies in a timely manner to avoid overfitting.

[0106] After training, use an independent test set to thoroughly examine the model's performance to ensure that it generalizes well and is applicable to unseen data.

[0107] The trained GCN model is applied to the actual test strip positive and negative identification task to assist in quickly and accurately judging the fault situation and improve diagnostic efficiency and service quality.

[0108] By combining the feature extraction capability of atrous convolution with a customized loss function, we not only improve the learning efficiency and accuracy of the model, but also enhance its understanding of complex failure modes, especially those with subtle differences.

[0109] S160. Use the graph convolutional neural network model to analyze the real-time data of the test strip to be tested, predict possible faults and their probabilities, and evaluate the color changes of the C line and T line of the test strip to preliminarily determine the positive or negative nature of the test strip.

[0110] In this embodiment, first, high-quality image data needs to be collected from the test strip. These images should include C line (control line) and T line (test line), and cover the entire reaction area as much as possible. The use of multi-spectral imaging equipment can obtain reflectivity information at different wavelengths, providing a richer data basis for subsequent color change analysis.

[0111] The iterative training process of the graph convolutional neural network is to use the prepared failure response dataset to perform end-to-end iterative training on the designed GCN model. The "failure response dataset" here actually refers to the test strip images under various failure modes collected and annotated from the historical data accumulated in the past. Therefore, this process is essentially training a GCN model based on historical data.

[0112] Based on the GCN model trained with historical data, the current state of the test strip is evaluated to predict the possible fault types and their probability of occurrence. For example, if the T line fails to change color as expected, it may be caused by insufficient samples or other problems; if the C line does not appear, it indicates that the test strip itself may be defective.

[0113] Color change assessment

[0114] Analyze the color changes of the C-line and T-line, and compare the standard reference values ​​to determine whether the actual test results meet expectations. This step not only focuses on the final color performance, but also considers the color change trend over time to improve the accuracy of judgment.

[0115] Based on all the above analysis results, a preliminary judgment on the positive or negative of the test strip is made. For qualitative testing, a positive usually means the presence of the target substance, while a negative means the absence. However, in some cases, further manual confirmation or laboratory testing is required to verify the results.

[0116] To achieve this process, a dedicated software tool or platform can be developed that integrates advanced image recognition algorithms and machine learning frameworks. Users only need to upload the image of the test strip to be tested, and the system can automatically complete the analysis and give a detailed report. In addition, a user-friendly interface can be designed so that non-professionals can easily operate and understand the analysis results.

[0117] By applying the graph convolutional neural network model, not only can the real-time data of the test strips be analyzed efficiently and accurately, but potential problems can also be warned in advance, helping users make diagnostic decisions faster and more accurately. This method improves the speed and reliability of medical diagnosis and has important practical application value.

[0118] S170. If the test result shows a positive result or there are other abnormal conditions, the color difference matrix and time series analysis method are used in combination with the multispectral image data to accurately distinguish different types of red, and the threshold judgment condition of the reflectance coefficient is automatically adjusted according to the sample characteristics to determine high-precision color change information.

[0119] In one embodiment, the above-mentioned step S170 may include steps S171 - S174 .

[0120] S171. Calibrate and standardize the multispectral image to obtain a standardized result.

[0121] In this embodiment, the collected multispectral images are subjected to preprocessing steps such as radiation correction and geometric correction to eliminate the influence of factors such as illumination changes and angle differences, thereby ensuring the comparability of data at different time points.

[0122] S172, performing feature extraction and color space conversion on the standardized result to construct a color difference matrix.

[0123] In this embodiment, the average reflectance at the pixel level is calculated for each band to form a multi-dimensional feature vector representing the color attribute of each pixel. A color difference matrix is ​​constructed, which records the difference in reflectance between two different bands under all possible combinations. This helps to highlight the subtle differences between different types of red.

[0124] Convert the original reflectance value to a space more suitable for color analysis, such as the CIELAB color space, where L represents brightness, a and b* represent the change axis from green to red and from blue to yellow respectively. This allows for a more intuitive understanding of the relative relationship between colors.

[0125] S173. Perform a time series analysis on the standardization result to obtain a time series analysis result.

[0126] In this embodiment, a time series analysis algorithm (such as dynamic time warping DTW, long short-term memory network LSTM) is applied to capture the color evolution trend over time, especially focusing on those rapid or gradual color changes that occur within a specific time period due to chemical reactions.

[0127] By combining historical data to set a baseline, an alarm is triggered when a new observed color change deviates from the normal range. This approach can help identify potential problems or special fault conditions.

[0128] S174, making a final color classification according to the color difference matrix, the timing analysis result and the threshold condition after adaptive adjustment to obtain high-precision color change information.

[0129] In this embodiment, a large number of known types of red samples are analyzed, and a model is established through machine learning or statistical methods to understand the typical manifestations of various types of red under different conditions. For example, some materials may show a deeper red when humidity increases.

[0130] Based on the above learning results, the initial reflectance thresholds are defined for each type of red. Then, in actual applications, these thresholds are dynamically adjusted according to the color changes monitored in real time, ensuring that high-precision color classification capabilities can be maintained even when environmental factors fluctuate.

[0131] Regularly retrain the model using the latest experimental data to continuously improve its adaptability to newly emerging color variations or complex scenes.

[0132] The final color classification decision is made by considering the results of the color difference matrix and timing analysis, as well as the threshold conditions after adaptive adjustment. This process should minimize the possibility of misjudgment while maximizing the correct recognition rate. Develop a user interface to simplify complex multispectral data analysis into an easy-to-understand form, such as a chart or color-coded map, to help operators quickly understand and respond to the detected color change information.

[0133] By integrating the color difference matrix and time series analysis methods, combined with multispectral image data and adaptive threshold adjustment strategies, the ability to distinguish different types of red can be significantly improved, thereby more accurately capturing subtle color changes, which is crucial for applications that require high sensitivity and precision (such as test strip analysis in medical diagnosis).

[0134] In this embodiment, the color difference matrix is ​​a tool used to quantify the reflectance differences between different bands in multispectral image analysis. It calculates the reflectance differences under all possible combinations between the bands and forms a matrix that records these differences.

[0135] For each pair of different bands, each element in the color difference matrix represents the difference in reflectance of the pair of bands at the same pixel location. This helps highlight subtle differences between different types of colors, especially in multispectral imaging, where different wavelengths of light can reveal color variations that are invisible to the naked eye.

[0136] The color difference matrix can more effectively capture the subtle differences in the color changes of the C line and T line of the test strip, especially for those color changes that are difficult to distinguish in the standard RGB color space.

[0137] Converting the raw reflectance values ​​to a space more suitable for color analysis (such as CIELAB) can further enhance the understanding and interpretation of color changes and make the relative relationship between colors more intuitive.

[0138] Time series analysis refers to the use of specific algorithms (such as dynamic time warping DTW, long short-term memory network LSTM, etc.) for standardized multispectral image data to study the color evolution trend over time.

[0139] This analysis focuses specifically on rapid or gradual color shifts over a specific period of time caused by chemical reactions or other factors, aiming to capture patterns of color change over time and identify anomalies or fault signatures.

[0140] Time series analysis can help monitor the color changes on the test strip over time, especially for those cases where the color changes occur due to chemical reactions, and can provide important information about the reaction rate, extent, etc.

[0141] By combining historical data to set a baseline and triggering an alarm when a newly observed color change deviates from the normal range, this approach helps to detect potential problems or special fault conditions at an early stage and improve the accuracy and reliability of diagnosis.

[0142] The color difference matrix and the results of timing analysis work together to provide a solid foundation for the final color classification. The color difference matrix provides static color contrast information, while timing analysis adds an understanding of dynamic changes in the time dimension. The combination of the two can not only better capture subtle color changes, but also predict possible faults and their probabilities, and evaluate the color changes of the C line and T line of the test strip, so as to preliminarily determine the positive and negative nature of the test strip. In addition, by adaptively adjusting the threshold conditions, the system can maintain high-precision color classification capabilities under fluctuating environmental factors, ensuring accurate and reliable detection results even in complex application scenarios.

[0143] High-precision color change information essentially refers to the detailed data of color properties and their changes over time or conditions obtained through precise measurement and analysis in specific application scenarios. This information not only covers the basic characteristics of color (such as hue, saturation, brightness), but also includes the quantitative description of the subtle differences between colors and the trend of color evolution over time.

[0144] In the application of medical diagnostic test strips, advanced methods such as multispectral imaging technology and graph convolutional neural network models are used to measure the color of the C line (control line) and T line (test line) on the test strip with ultra-high precision. This includes but is not limited to the spatial distribution of color, color intensity, hue, saturation and brightness. The time-varying color evolution trend is captured by the time series analysis algorithm, especially those rapid or gradual color changes caused by chemical reactions. This method can help identify potential problems or special failure conditions, such as whether the test strip has correctly undergone the expected chemical reaction. Constructing a color difference matrix to quantify the difference in reflectance between different bands helps to highlight the subtle differences between different types of colors, especially those changes that are difficult to detect with the naked eye. This allows the system to more effectively distinguish between background noise and meaningful color change signals. Models based on machine learning or statistical methods understand the typical manifestations of various types of red (or other related colors) under different conditions and define initial reflectance coefficient thresholds for each type of red. These thresholds are then dynamically adjusted based on the color changes monitored in real time to ensure that high-precision color classification capabilities can be maintained even when environmental factors fluctuate. The final color classification decision is made by considering the results of the color difference matrix and the timing analysis. This process should minimize the possibility of misjudgment while maximizing the correct recognition rate. Develop a user interface that simplifies complex multispectral data analysis into an easy-to-understand form to help operators quickly understand and respond to detected color change information.

[0145] High-precision color change information means that extremely high resolution and accuracy are achieved when processing color-related data, not only to see the color changes on the surface, but also to deeply understand the physical or chemical processes behind these changes. This is crucial for applications that require high sensitivity and precision (such as test strip analysis in medical diagnosis) because it directly affects the reliability and effectiveness of the diagnostic results.

[0146] S180. Based on the optimized feature data and the high-precision color change information, the confidence is calculated, including the coordinates of the object frame, the object confidence, the category of the object frame and the category confidence, the severity level of the object frame and the severity level confidence, and the object frame is filtered and post-processed through non-maximum suppression and other techniques to obtain the target object frame.

[0147] In one embodiment, the above-mentioned step S180 may include steps S181 - S182 .

[0148] S181, fusing the optimized feature data and the high-precision color change information to obtain a fusion result;

[0149] S182. Calculate the confidence of the fusion result, including the coordinates of the object frame, the object confidence, the category of the object frame and the category confidence, the severity level of the object frame and the severity level confidence, and filter and post-process the object frame through non-maximum suppression and other technologies to obtain the target object frame.

[0150] In this embodiment, the optimized feature data (such as reflectance and color difference matrix results) extracted from the multispectral image are combined with the high-precision color change information obtained through time series analysis. This step ensures that the data used for subsequent analysis contains both the characteristics in the spatial dimension and the dynamic changes in the temporal dimension.

[0151] Using machine learning or deep learning algorithms (such as convolutional neural networks (CNNs)), the region of interest (ROI) in the image is identified based on the fused feature data, and a bounding box (i.e., object box) is generated for each detected object. This process outputs the specific coordinate position of each object box.

[0152] For each predicted object box, calculate the probability of its existence, that is, the object confidence. This is usually directly output by the model and reflects the model's confidence level that the target object does exist in the object box.

[0153] The trained classifier is further used to determine which category the object in the object box belongs to (for example, different types of red), and the category confidence is given, indicating how confident the model is about the classification result.

[0154] According to predefined standards or rule sets, each object box is assigned a severity level (such as mild, moderate, severe) in combination with the color change information and other relevant features of the object. Similarly, the corresponding severity level confidence is also required to measure the reliability of this assessment.

[0155] Implement a non-maximum suppression algorithm to remove redundant object boxes with high overlap and low confidence, and retain the boxes that are most likely to contain actual targets. This helps reduce false positives and improve detection accuracy.

[0156] After NMS, additional post-processing steps are required, such as:

[0157] Boundary Smoothing: Fine-tune the edges of the object box to make it fit the target shape better.

[0158] Size Limit: Set minimum / maximum size thresholds to exclude objects of abnormal size.

[0159] Contextual verification: Consider the surrounding environment information to confirm whether the object box is reasonable.

[0160] After all the above processing, the target object boxes that meet the standards are screened out as the final output. These boxes should have high confidence, clear category labels, reasonable severity levels, and have been fully denoised and smoothed.

[0161] The finalized target object box and its related information (coordinates, confidence, category, severity level, etc.) are organized into a structured format for further application or visualization to the user. In addition, a graphical interface can be developed to intuitively display these detection results to help decision makers quickly understand the situation and take appropriate actions.

[0162] In summary, this process integrates advanced image processing technology and intelligent analysis methods, which can not only accurately locate and identify specific objects in images, but also make detailed evaluations of their properties, thus providing strong support for various application scenarios.

[0163] In this embodiment, when the trained graph convolutional neural network (GCN) model is used to analyze the real-time data of the test strip to be tested, possible faults and their probabilities are predicted, and the color changes of the C line and T line of the test strip are evaluated to preliminarily determine the positive or negative of the test strip. This means that a preliminary positive or negative test result can be obtained at this step.

[0164] Next, in S170, if the test result is positive or there are other abnormal conditions, the color difference matrix and time series analysis method are used to accurately distinguish different types of red in combination with multispectral image data, and the threshold judgment condition of the reflectance coefficient is automatically adjusted according to the sample characteristics to determine high-precision color change information. This step is intended to improve the accuracy and sensitivity of color change recognition, thereby indirectly enhancing the accuracy of positive and negative judgment of the test strip.

[0165] Finally, in S180, based on the optimized feature data and high-precision color change information, the confidence is calculated, including the coordinates of the object frame, the object confidence, the category of the object frame and the category confidence, the severity level of the object frame and the severity level confidence. The object frame is filtered and post-processed through non-maximum suppression and other technologies to obtain the target object frame, and the final confidence is calculated based on this to assist in diagnosis and decision-making. Therefore, the final positive and negative detection results and their confidence are determined here.

[0166] In summary, the preliminary test result of the positive or negative of the test strip is obtained in S160, and the final test result and its reliability (confidence) are determined in S180. As an intermediate link, S170 enhances the accuracy of color change recognition, thereby improving the reliability of the final positive or negative judgment.

[0167] The above-mentioned method for improving the accuracy of positive and negative test strip recognition collects multispectral images by combining a high-resolution camera with red, green and near-infrared light sources to enhance the richness and reliability of image data; uses multispectral image data for feature extraction and optimization to generate optimized feature vectors to improve the accuracy of model recognition; constructs a fault injection test method based on a logic circuit model to simulate problems in the actual environment by randomly injecting faults to enhance the robustness of the model; uses a graph convolutional neural network for iterative training, and improves the feature extraction capability of the model under different fault modes through void convolution; accurately distinguishes the type of red through color difference matrix and timing analysis, and automatically adjusts the judgment conditions to further improve the accuracy of color change information and enhance the model's ability to accurately distinguish test strips.

[0168] Figure 2 is a schematic block diagram of a device 300 for improving the accuracy of positive / negative identification of a test strip provided by an embodiment of the present invention. Figure 2 As shown, corresponding to the above test strip positive and negative identification accuracy improvement method, the present invention also provides a test strip positive and negative identification accuracy improvement device 300. The test strip positive and negative identification accuracy improvement device 300 includes a unit for executing the above test strip positive and negative identification accuracy improvement method, and the device can be configured in a server. Specifically, please refer to Figure 2 The device 300 for improving the accuracy of positive and negative identification of test strips includes a data acquisition unit 301, an optimization unit 302, a model abstraction unit 303, a fault simulation unit 304, an iterative training unit 305, a fault prediction unit 306, an analysis unit 307 and a final prediction unit 308.

[0169] The data acquisition unit 301 is used to use a high-resolution camera in combination with three wavelength light sources of red light, green light and near-infrared light to perform multispectral signal acquisition on the C line, T line and the area between the test strips to obtain multispectral image data; the optimization unit 302 is used to extract and optimize the multispectral image data to obtain an optimized feature vector; the model abstraction unit 303 is used to abstract the test strip detection process into a logic circuit model based on the optimized feature vector, wherein the key operation steps are used as nodes, the input-output relationship constitutes an edge, and a node feature matrix is ​​constructed. The logic circuit model is used to guide fault injection testing and training data generation; the fault simulation unit 304 is used to randomly inject faults at specific node positions in the logic circuit model to simulate problems that may occur in a real environment, generate corresponding test vectors based on the logic circuit model and record failure response data; the iterative training unit 305 is used to iteratively train the graph convolutional neural network model using the failure response data, and add hole convolutions of different sizes in multiple network layers to increase Strong feature extraction capability enables the model to more effectively learn the feature representations under different fault modes, and finally obtain a trained graph convolutional neural network model; a fault prediction unit 306, which is used to apply the graph convolutional neural network model to analyze the real-time data of the test strip to be tested, predict possible faults and their probabilities, and evaluate the color changes of the C line and T line of the test strip to preliminarily judge the positive and negative nature of the test strip; an analysis unit 307, which is used to use the color difference matrix and time series analysis method to accurately distinguish different types of red in combination with the multispectral image data if the test result shows a positive or other abnormal conditions exist, and automatically adjust the threshold judgment condition of the reflectance coefficient according to the sample characteristics to determine high-precision color change information; a final prediction unit 308, which is used to calculate the confidence based on the optimized feature data and the high-precision color change information, including the coordinates of the object frame, the object confidence, the category and category confidence of the object frame, the severity level and severity level confidence of the object frame, and filter and post-process the object frame through non-maximum suppression and other technologies to obtain the target object frame.

[0170] In one embodiment, the optimization unit 302 is used to:

[0171] The multispectral image data is preprocessed to obtain a preprocessing result; the preprocessing result is extracted based on color spatial features, set features and texture features to obtain an extraction result; the extraction result is dimensionality reduced, feature selected and feature enhanced to obtain an optimization result; a feature vector is constructed according to the optimization result to obtain an optimized feature vector.

[0172] In one embodiment, the fault simulation unit 304 is used to:

[0173] Determine key nodes and fault modes for the logic circuit model; determine a fault injection strategy; inject faults at the key nodes in combination with the fault modes according to the fault injection strategy, and generate a test vector containing current fault information and its impact range after each fault injection is completed; and record failure response data according to the test vector.

[0174] In one embodiment, the iterative training unit 305 is used to:

[0175] Preprocessing the failure response data to obtain a processing result; constructing a multi-layer GCN model; and training the multi-layer GCN model using the processing result to obtain a trained graph convolutional neural network model.

[0176] In one embodiment, the analysis unit 307 is used to:

[0177] The multispectral image is calibrated and standardized to obtain a standardized result; the standardized result is subjected to feature extraction and color space conversion to construct a color difference matrix; the standardized result is subjected to time series analysis to obtain a time series analysis result; and a final color classification is performed based on the color difference matrix, the time series analysis result and the threshold condition after adaptive adjustment to obtain high-precision color change information.

[0178] In one embodiment, the final prediction unit 308 is used to:

[0179] The optimized feature data and the high-precision color change information are fused to obtain a fusion result; the confidence of the fusion result is calculated, including the coordinates of the object frame, the object confidence, the category of the object frame and the category confidence, the severity level of the object frame and the severity level confidence, and the object frame is filtered and post-processed through non-maximum suppression and other technologies to obtain a target object frame.

[0180] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned test strip positive / negative identification accuracy improvement device 300 and each unit can refer to the corresponding description in the aforementioned method embodiment, and for the convenience and brevity of description, it will not be repeated here.

[0181] The above-mentioned device 300 for improving the accuracy of positive / negative identification of test strips can be implemented in the form of a computer program. The computer program can be used in Figure 3 Runs on the computer device shown.

[0182] See also Figure 3 , Figure 35 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device 500 may be a server, wherein the server may be an independent server or a server cluster composed of multiple servers.

[0183] See also Figure 3 The computer device 500 includes a processor 502 , a memory and a network interface 505 connected via a system bus 501 , wherein the memory may include a non-volatile storage medium 503 and an internal memory 504 .

[0184] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions, and when the program instructions are executed, the processor 502 can execute a method for improving the accuracy of positive and negative identification of a test strip.

[0185] The processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500 .

[0186] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a method for improving the accuracy of positive and negative identification of test strips.

[0187] The network interface 505 is used to communicate with other devices over the network. Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0188] The processor 502 is used to run the computer program 5032 stored in the memory to implement the following steps:

[0189] A high-resolution camera is used in combination with three wavelength light sources of red light, green light and near-infrared light to collect multispectral signals of the C line, T line and the area between the test strips to obtain multispectral image data; the multispectral image data is feature extracted and optimized to obtain an optimized feature vector; based on the optimized feature vector, the test strip detection process is abstracted into a logic circuit model, in which key operation steps are used as nodes, input-output relationships constitute edges, and a node feature matrix is ​​constructed. The logic circuit model is used to guide fault injection testing and training data generation; faults are randomly injected into specific node positions in the logic circuit model to simulate problems that may occur in a real environment, and corresponding test vectors are generated based on the logic circuit model and failure response data are recorded; the graph convolutional neural network model is iteratively trained using the failure response data, and different sizes of hole convolutions are added to multiple network layers to enhance feature extraction capabilities, so that the model can The feature representations under different fault modes can be learned more effectively, and finally a trained graph convolutional neural network model is obtained; the graph convolutional neural network model is used to analyze the real-time data of the test strip to be tested, predict possible faults and their probabilities, and evaluate the color changes of the C line and T line of the test strip to preliminarily judge the positive and negative nature of the test strip; if the test result shows a positive or there are other abnormal conditions, the color difference matrix and time series analysis method are used to combine the multispectral image data to accurately distinguish different types of red, and the threshold judgment condition of the reflectance coefficient is automatically adjusted according to the sample characteristics to determine high-precision color change information; based on the optimized feature data and the high-precision color change information, the confidence is calculated, including the coordinates of the object frame, the object confidence, the category and category confidence of the object frame, the severity level and severity level confidence of the object frame, and the object frame is filtered and post-processed by non-maximum suppression and other technologies to obtain the target object frame.

[0190] It should be understood that in the embodiment of the present application, the processor 502 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0191] It can be understood by those skilled in the art that all or part of the processes in the method for implementing the above embodiment can be completed by instructing the relevant hardware through a computer program. The computer program includes program instructions, and the computer program can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiment of the above method.

[0192] Therefore, the present invention also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein when the computer program is executed by a processor, the processor executes the following steps:

[0193] A high-resolution camera is used in combination with three wavelength light sources of red light, green light and near-infrared light to collect multispectral signals of the C line, T line and the area between the test strips to obtain multispectral image data; the multispectral image data is feature extracted and optimized to obtain an optimized feature vector; based on the optimized feature vector, the test strip detection process is abstracted into a logic circuit model, in which key operation steps are used as nodes, input-output relationships constitute edges, and a node feature matrix is ​​constructed. The logic circuit model is used to guide fault injection testing and training data generation; faults are randomly injected into specific node positions in the logic circuit model to simulate problems that may occur in a real environment, and corresponding test vectors are generated based on the logic circuit model and failure response data are recorded; the graph convolutional neural network model is iteratively trained using the failure response data, and different sizes of hole convolutions are added to multiple network layers to enhance feature extraction capabilities, so that the model can The feature representations under different fault modes can be learned more effectively, and finally a trained graph convolutional neural network model is obtained; the graph convolutional neural network model is used to analyze the real-time data of the test strip to be tested, predict possible faults and their probabilities, and evaluate the color changes of the C line and T line of the test strip to preliminarily judge the positive and negative nature of the test strip; if the test result shows a positive or there are other abnormal conditions, the color difference matrix and time series analysis method are used to combine the multispectral image data to accurately distinguish different types of red, and the threshold judgment condition of the reflectance coefficient is automatically adjusted according to the sample characteristics to determine high-precision color change information; based on the optimized feature data and the high-precision color change information, the confidence is calculated, including the coordinates of the object frame, the object confidence, the category and category confidence of the object frame, the severity level and severity level confidence of the object frame, and the object frame is filtered and post-processed by non-maximum suppression and other technologies to obtain the target object frame.

[0194] The storage medium may be a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk, etc., which are computer-readable storage media that can store program codes.

[0195] It should be noted that the functions or steps that can be implemented by the above storage medium or computer device can refer to the relevant descriptions on the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0196] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0197] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of each unit is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0198] The steps in the method of the embodiment of the present invention can be adjusted in order, combined and deleted according to actual needs. The units in the device of the embodiment of the present invention can be combined, divided and deleted according to actual needs. In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0199] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, terminal, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.

[0200] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for improving the accuracy of positive and negative identification of test strips, characterized in that: include: A high-resolution camera is used in combination with three wavelength light sources of red light, green light and near-infrared light to collect multispectral signals of the C line, T line and the area between the test strips to obtain multispectral image data; Extracting and optimizing the multispectral image data to obtain an optimized feature vector; Based on the optimized feature vector, the test strip detection process is abstracted into a logic circuit model, wherein the key operation steps are used as nodes, the input-output relationship constitutes the edge, and the node feature matrix is ​​constructed. The logic circuit model is used to guide fault injection testing and training data generation; Randomly injecting faults into key node positions in the logic circuit model to simulate problems that occur in a real environment, generating corresponding test vectors based on the logic circuit model and recording failure response data; The graph convolutional neural network model is iteratively trained using the failure response data, and hole convolutions of different sizes are added to multiple network layers to enhance feature extraction capabilities, so that the model can more effectively learn feature representations under different fault modes, and finally a trained graph convolutional neural network model is obtained; The graph convolutional neural network model is used to analyze the real-time data of the test strips to be tested, predict the existing faults and their probabilities, and evaluate the color changes of the C line and T line of the test strips to preliminarily determine the positive or negative of the test strips; If the test result shows positive, the color difference matrix and time series analysis method are used to accurately distinguish different types of red in combination with the multispectral image data, and the threshold judgment condition of the reflectance coefficient is automatically adjusted according to the sample characteristics to determine high-precision color change information; Based on the optimized feature data and the high-precision color change information, the confidence is calculated, including the coordinates of the object frame, the object confidence, the category of the object frame and the category confidence, the severity level of the object frame and the severity level confidence, and the object frame is filtered and post-processed by non-maximum suppression technology to obtain a target object frame; Using machine learning or deep learning algorithms, the region of interest in the image is identified based on the fused feature data, and a bounding box is generated for each detected object, and the specific coordinate position of each object box is output; the object box refers to the area where the C line and T line are located; For each predicted object box, calculate the probability of existence to form the object confidence; The trained classifier is further used to determine the category to which the object in the object box belongs, where the category refers to different types of red, and the category confidence is given; According to predefined standards or rule sets, combined with the color change information and features of the object, a severity level is assigned to each object frame, and the corresponding severity level confidence is output; the severity levels include mild, medium, and severe.

2. The method for improving the accuracy of positive and negative identification of a test strip according to claim 1, characterized in that: The extracting and optimizing the multispectral image data to obtain an optimized feature vector includes: Preprocessing the multispectral image data to obtain a preprocessing result; Extracting the preprocessing result based on color spatial features, set features and texture features to obtain an extraction result; Performing dimensionality reduction, feature selection, and feature enhancement on the extraction results to obtain an optimized result; A feature vector is constructed according to the optimization result to obtain an optimized feature vector.

3. The method for improving the accuracy of positive and negative identification of a test strip according to claim 2, characterized in that: The randomly injecting faults into the key node positions in the logic circuit model to simulate the problems occurring in the real environment, generating corresponding test vectors based on the logic circuit model and recording failure response data, includes: Determining key nodes and failure modes for the logic circuit model; Determine the fault injection strategy; Injecting faults at the key nodes in combination with fault modes according to the fault injection strategy, and generating a test vector including current fault information and its impact range after each fault injection is completed; Failure response data is recorded according to the test vector.

4. The method for improving the accuracy of positive and negative identification of a test strip according to claim 3, characterized in that: The graph convolutional neural network model is iteratively trained using the failure response data, and different sizes of hole convolutions are added to multiple network layers to enhance feature extraction capabilities, so that the model can more effectively learn feature representations under different fault modes, and finally a trained graph convolutional neural network model is obtained, including: Preprocessing the failure response data to obtain a processing result; Build a multi-layer GCN model; The processing results are used to train a multi-layer GCN model to obtain a trained graph convolutional neural network model.

5. The method for improving the accuracy of positive / negative identification of a test strip according to claim 4, characterized in that: Each layer of the multi-layer GCN model contains standard GCN operations for aggregating neighbor information, as well as additional dilated convolution layers to capture contextual information in a wider range without significantly increasing the number of parameters. Dilated convolutions are introduced at different levels of the multi-layer GCN model, and the expansion rate is adjusted to enable the model to perceive features of different scales. Customized loss functions are used in the model training process, including object box loss calculation function, object loss function, category loss function and severity loss function. The severity loss function adds a sample probability penalty term on the basis of the cross entropy loss function to improve the model's sensitivity to slight changes.

6. The method for improving the accuracy of positive and negative identification of a test strip according to claim 1, characterized in that: The color difference matrix and time series analysis method are used to accurately distinguish different types of red in combination with the multi-spectral image data, and the threshold judgment condition of the reflectance coefficient is automatically adjusted according to the sample characteristics to determine high-precision color change information, including: Calibrate and standardize the multispectral image to obtain a standardized result; Performing feature extraction and color space conversion on the standardized result to construct a color difference matrix; Performing time series analysis on the standardized result to obtain a time series analysis result; The final color classification is made according to the color difference matrix, the timing analysis result and the threshold condition after adaptive adjustment to obtain high-precision color change information.

7. The method for improving the accuracy of positive and negative identification of a test strip according to claim 1, characterized in that: The confidence is calculated based on the optimized feature data and the high-precision color change information, including the coordinates of the object frame, the object confidence, the category of the object frame and the category confidence, the severity level of the object frame and the severity level confidence, and the object frame is filtered and post-processed by non-maximum suppression technology to obtain the target object frame, including: Fusing the optimized feature data and the high-precision color change information to obtain a fusion result; The confidence of the fusion result is calculated, including the coordinates of the object frame, the object confidence, the category of the object frame and the category confidence, the severity level of the object frame and the severity level confidence, and the object frame is filtered and post-processed by non-maximum suppression technology to obtain the target object frame.

8. A device for improving the accuracy of positive and negative identification of test strips, characterized in that: include: A data acquisition unit is used to collect multispectral signals of the C line, T line and the area between the test strips using a high-resolution camera combined with three wavelength light sources of red light, green light and near-infrared light to obtain multispectral image data; An optimization unit, used for extracting and optimizing features of the multispectral image data to obtain an optimized feature vector; A model abstraction unit, used to abstract the test strip detection process into a logic circuit model based on the optimized feature vector, wherein the key operation steps are used as nodes, the input-output relationship constitutes an edge, and a node feature matrix is ​​constructed. The logic circuit model is used to guide fault injection testing and training data generation; A fault simulation unit, used for randomly injecting faults into key node positions in the logic circuit model to simulate problems occurring in a real environment, generating corresponding test vectors based on the logic circuit model and recording failure response data; An iterative training unit, used to iteratively train the graph convolutional neural network model using the failure response data, adding hole convolutions of different sizes in multiple network layers to enhance feature extraction capabilities, so that the model can more effectively learn feature representations under different fault modes, and finally obtain a trained graph convolutional neural network model; The fault prediction unit is used to apply the graph convolutional neural network model to analyze the real-time data of the test strip to be tested, predict the existing faults and their probabilities, and evaluate the color changes of the C line and T line of the test strip to preliminarily judge the positive or negative of the test strip; An analysis unit, for, if the test result shows a positive result, using a color difference matrix and a time series analysis method in combination with the multispectral image data to accurately distinguish different types of red, and automatically adjusting a threshold judgment condition of the reflectance coefficient according to sample characteristics to determine high-precision color change information; A final prediction unit is used to calculate the confidence, including the coordinates of the object frame, the object confidence, the category and category confidence of the object frame, the severity level and severity level confidence of the object frame based on the optimized feature data and the high-precision color change information, and filter and post-process the object frame through non-maximum suppression technology to obtain a target object frame; Using machine learning or deep learning algorithms, the region of interest in the image is identified based on the fused feature data, and a bounding box is generated for each detected object, and the specific coordinate position of each object box is output; the object box refers to the area where the C line and T line are located; For each predicted object box, calculate the probability of existence to form the object confidence; The trained classifier is further used to determine the category to which the object in the object box belongs, where the category refers to different types of red, and the category confidence is given; According to predefined standards or rule sets, combined with the color change information and features of the object, a severity level is assigned to each object frame, and the corresponding severity level confidence is output; the severity levels include mild, medium, and severe.

9. A computer device, characterized in that: The computer device comprises a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images