A medical waste whole-process-oriented traceability and risk early warning method and system

Abstract

The present application relates to the field of data processing, and particularly relates to a medical waste whole-process traceability and risk early warning method and system. The method comprises the following steps: generating loss labels and constructing matching pairs; determining feature weights of corresponding features of a pre-training neural network; obtaining a predicted category of each loss label; comparing the predicted category of the loss label with a historical category of an unscanned transfer label; if the categories are consistent, calculating a current weight difference between each known label and the loss label, and a historical weight difference between each known label and the unscanned transfer label, and calculating a preliminary matching measure; generating a weight penalty term; fusing the preliminary matching measure and the weight penalty term to generate a final matching cost of each matching pair for global optimal matching, and determining whole-process traceability information for the loss label that is successfully matched. The present application can improve the robustness of medical waste whole-process traceability and reduce the cost of manual verification.

Description

A method and system for traceability and risk warning of medical waste throughout the entire process. Technical Field

[0001] This invention relates to the field of data processing, specifically to a method and system for tracing and risk warning of the entire process of medical waste. Background Technology

[0002] Medical waste refers to waste generated during medical, preventive, healthcare, and related activities that possesses direct or indirect infectiousness, toxicity, or other hazards. This type of waste is highly hazardous, and risks such as loss, leakage, or illegal dumping can occur at every stage from generation, sorting, collection, temporary storage, and transfer to final disposal. Implementing end-to-end traceability allows for real-time tracking of the flow of each bag of waste, ensuring compliant disposal and identifying potential problems such as abnormal handovers, sudden weight changes, or disruptions in the timeline, enabling timely warnings and preventing medical waste from entering illegal channels or polluting the environment.

[0003] Currently, a typical medical waste traceability and risk warning system usually adopts the following closed-loop process: At the source collection and labeling stage, medical staff complete waste sorting and packaging, then use specialized equipment to automatically weigh and generate and affix QR code labels. Upon entering the transportation stage, collection personnel complete the handover by scanning the QR code, and the intelligent transport vehicle transports the waste along a planned route to the temporary storage room, recording the operators, time, and route information throughout the process. At the out-of-hospital handover stage, disposal company personnel scan the medical waste packaging and its outbound QR code to complete the electronic manifest handover. After the waste arrives at the disposal center, it is scanned again for verification and final disposal is performed, thus forming a complete data closed loop. Throughout the entire process, the system continuously monitors risky behaviors such as abnormal weight, handover delays, route deviations, and video violations. Once a preset warning rule is triggered, an alarm message is immediately pushed.

[0004] However, existing medical waste management systems often face challenges in label identification due to the inherent characteristics of the waste and the unique environment in which it is stored. Medical waste is frequently contaminated with blood, bodily fluids, and chemical disinfectants, and possesses physical characteristics such as high humidity and the presence of sharp objects. Furthermore, waste is categorized into five main types: infectious, pathological, damaging, pharmaceutical, and chemical, resulting in significant differences in packaging materials, sealing methods, and storage conditions. This diversity makes traditional QR code labels highly susceptible to damage during transport, temporary storage, and handover from liquid penetration, chemical corrosion, scratches from sharp objects, or surface contamination. This can lead to blurred, damaged, or detached barcodes, causing scanning failures and information chain breaks in subsequent stages. This severely undermines the credibility and data integrity of the entire traceability process, exacerbating management loopholes and potential risks. Summary of the Invention

[0005] This invention provides a method and system for tracing and risk warning of medical waste throughout the entire process, in order to solve existing problems.

[0006] The present invention provides a method for tracing and risk warning of medical waste throughout the entire process, which adopts the following technical solution:

[0007] One embodiment of the present invention provides a method for traceability and risk warning of medical waste throughout the entire process, the method comprising the following steps:

[0008] At the handover node, a loss tag is generated for medical waste packages that cannot be scanned, and a matching pair is formed with the unscanned handover tag that has been recorded at the previous node but has not appeared in this node.

[0009] Test the pre-trained neural network using all known labels of this node, and determine the feature weights of the corresponding features based on the classification accuracy of each pre-trained neural network.

[0010] Based on this feature weight, cluster analysis is performed on all loss labels to obtain the predicted category of each loss label;

[0011] For each matching pair, compare the predicted category of the loss label with the historical category of the unscanned handover label in that matching pair;

[0012] If consistent, then using all known labels of this node as references, calculate the current weight difference between each known label and the lost label, and the historical weight difference between each known label and the unscanned overlapping label. Calculate the preliminary matching metric based on the consistency of the current weight difference and the historical weight difference. Obtain the weight stability weight according to the predicted category of the lost label, and weight the absolute weight difference between the two labels in the matching pair to generate a weight penalty term. Combine the preliminary matching metric and the weight penalty term to generate the final matching cost for each matching pair.

[0013] If they are inconsistent, a maximum cost is generated for each matching pair as the final matching cost;

[0014] A cost matrix is ​​constructed based on the final matching cost of all matching pairs. The Hungarian algorithm is used to perform global optimal matching, and the loss label of successful matching is used to determine the full-process traceability information.

[0015] Optionally, the pre-trained neural network is tested using all known labels of this node, and the feature weights of the corresponding features are determined based on the classification accuracy of each pre-trained neural network, specifically including:

[0016] Input the packaged images of all known labels of this node into each pre-trained neural network to obtain the classification results of each pre-trained neural network for each known label;

[0017] The classification result of each known label is compared with the true category of that known label;

[0018] Count the number of known labels correctly classified by each pre-trained neural network;

[0019] Based on the number of correctly classified known labels, feature weights are assigned to the features corresponding to each pre-trained neural network. The assigned feature weights are positively correlated with the number of correctly classified labels.

[0020] Optionally, cluster analysis is performed on all loss labels based on this feature weight to obtain the predicted category for each loss label, specifically including:

[0021] For any two loss labels, input the packaged images of these two loss labels into each pre-trained neural network respectively;

[0022] For each pre-trained neural network, determine whether its classification results for the two loss labels are the same;

[0023] If they are the same, the feature weights corresponding to the pre-trained neural network are normalized and included in the clustering contribution value of the two loss labels; if they are different, zero is included in the clustering contribution value.

[0024] The clustering contribution values ​​of the two loss labels are summed to obtain the similarity score between the two loss labels;

[0025] Based on the similarity scores between all pairs of loss labels, clustering algorithms are used to cluster all loss labels to obtain classification clusters;

[0026] Each classification cluster is matched with each predefined standard category of medical waste to determine the medical waste category corresponding to each classification cluster, and this medical waste category is determined as the predicted category for each loss label in that classification cluster.

[0027] Optionally, using all known labels of this node as a reference, calculate the current weight difference between each known label and the lost label, and the historical weight difference between each known label and the unscanned overlapping label. Based on the consistency between the current weight difference and the historical weight difference, calculate a preliminary matching metric, specifically including:

[0028] For each known tag in this node, calculate the first weight difference between its current weight and the current weight of the lost tag, and calculate the second weight difference between its historical weight in the previous node and the historical weight of the unscanned handover tag.

[0029] Calculate the absolute value of the difference between the first weight difference and the second weight difference;

[0030] The matching contribution of the known label to the matching pair is determined based on the magnitude of the absolute value, where the absolute value and the matching contribution are negatively correlated.

[0031] Calculate the absolute value of the difference between the current weight of the known tag and the current weight of the lost tag;

[0032] A weight is assigned to the known label based on the absolute value of the weight difference, wherein the absolute value of the weight difference and the assigned weight are negatively correlated.

[0033] The matching contribution value of the known label is weighted according to the assigned weights to obtain the weighted contribution of the known label;

[0034] Sum the weighted contributions of all known labels and use the summation result as a preliminary matching measure.

[0035] Optionally, a weight stability weight is obtained based on the predicted category of the loss label, and the absolute weight difference between the two labels in the matching pair is weighted to generate a weight penalty term, specifically including:

[0036] Based on the predicted category of the loss label, the corresponding weight stability weight is obtained by querying the preset weight stability weight mapping table. The preset weight stability weight mapping table is established based on historical data analysis, and the weight stability weight in the preset weight stability weight mapping table is positively correlated with the stability of weight change of medical waste category.

[0037] Calculate the absolute value of the difference between the current weight of the lost tag in the matching pair and the historical weight of the unscanned handover tag, and normalize it to obtain the absolute weight difference.

[0038] Multiply the absolute weight difference by the weight stability weight to obtain the weight penalty term.

[0039] Optionally, the final matching cost for each matching pair is generated by fusing the initial matching metric and the weight penalty term, specifically including:

[0040] The initial matching metric is converted into an initial matching cost, where the initial matching metric and the initial matching cost are negatively correlated.

[0041] Multiply the initial matching cost by the weight penalty term to obtain the final matching cost of the matching pair.

[0042] Optionally, a cost matrix is ​​constructed based on the final matching costs of all matching pairs. The Hungarian algorithm is then used to perform globally optimal matching, and full-process traceability information is determined for the loss labels of successful matches. This specifically includes:

[0043] The matrix is ​​composed of rows of all loss labels and columns of all unscanned intersection labels.

[0044] The final matching cost of each matching pair is filled into the matrix element position corresponding to the intersection of the row of the loss label and the column of the unscanned intersection label in that matching pair, to obtain the initial cost matrix;

[0045] If the number of lost labels is not equal to the number of unscanned handover labels, then add rows or columns filled with preset values ​​to the matrix to expand the initial cost matrix into a square matrix, thus obtaining the cost matrix.

[0046] The cost matrix is ​​input into the Hungarian algorithm for calculation, and the globally optimal matching scheme with the minimum total cost is obtained.

[0047] Based on the optimal matching scheme, determine the unscanned handover label that matches each lost label, and perform data association for each successfully matched pair of labels;

[0048] The complete historical traceability information of the unscanned handover label that was successfully matched from the node to the previous node is assigned to the matching loss label, and the information of the loss label of this node is completed to obtain the full-process traceability information of the loss label in this node.

[0049] For loss labels that fail to match successfully in the optimal matching scheme or for unscanned handover labels, generate a manual verification warning.

[0050] Optionally, feature weights are assigned to the features corresponding to each pre-trained neural network based on the number of correctly classified known labels, specifically including:

[0051] The classification accuracy of a network is obtained by dividing the number of known labels correctly classified by each pre-trained neural network by the total number of known labels of that node, and this classification accuracy is used as the feature weight of its corresponding feature.

[0052] Optionally, the clustering algorithm is the K-means clustering algorithm, and the clustering parameter of the clustering algorithm is set to 5.

[0053] This invention proposes a traceability and risk warning system for the entire process of medical waste, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the traceability and risk warning method for the entire process of medical waste.

[0054] The beneficial effects of the technical solution of the present invention are:

[0055] In this embodiment of the invention, by integrating multimodal perception and intelligent matching mechanisms, the digital identity and circulation chain of medical waste can still be reconstructed even when physical tags are damaged. It not only achieves real-time classification verification based on visual features, consistency analysis of circulation combined with temporal patterns, and weight-based fault tolerance judgment according to category differences, but also completes automatic repair and supplementation of cross-link data chains through a global optimization algorithm. This solution effectively solves the problem of information chain breakage caused by tag damage, improves the robustness and data integrity of the traceability system, reduces manual verification costs, and provides a reliable data foundation for risk warning throughout the entire process. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 is a flowchart of a method for tracing and risk warning of the entire medical waste process provided by an embodiment of the present invention;

[0058] Figure 2 is a structural diagram of a traceability and risk warning system for the entire process of medical waste provided by an embodiment of the present invention. Detailed Implementation

[0059] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method for tracing and risk warning of the entire medical waste process according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0060] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0061] The following description, in conjunction with the accompanying drawings, details a specific scheme for a method of tracing and risk warning for the entire process of medical waste provided by this invention.

[0062] This invention provides a method and system for tracing and risk warning of medical waste throughout the entire process. Please refer to Figure 1, which shows a flowchart of a method for tracing and risk warning of medical waste throughout the entire process according to an embodiment of this invention. The method includes the following steps:

[0063] S101. At the handover node, generate a loss tag for medical waste packages that cannot be scanned, and form a matching pair with the unscanned handover tags that have been recorded in the previous node but have not appeared in this node.

[0064] For example, at each standard handover point for medical waste (e.g., the transfer point from the medical department to the hospital's temporary storage room), when an operator attempts to scan the QR code on the medical waste packaging for handover registration, if the label is contaminated, damaged, or completely detached and cannot be read, the system will automatically execute the following processing procedure:

[0065] Generate a loss tag: The system uses industrial cameras to capture multi-angle images of the unidentifiable waste packaging; it obtains real-time weight data of the waste package from an integrated smart weighing platform; and it automatically records the timestamp of the event, the GPS or beacon location of the node, and the operator's employee number. Subsequently, the system creates a new record in the "Current Node Anomaly Table" of the traceability database, which is marked as a "loss tag." It is important to note that the loss tag only contains the instantaneous information collected at the current node; it does not have a unique identification code (ID) associated with any historical handover records. Its purpose is to represent a waste package object where "the physical entity has arrived, but its electronic identity has been lost."

[0066] Linking unscanned handover tags: At the same time, the system background accesses the "Normal Flow Record Table" in the traceability database and executes a query. The query logic is: to find all waste package records that have been successfully scanned and registered at the next adjacent handover node (such as the preceding department), and whose task destination points to this node, but have not yet been scanned and confirmed by this node within the expected time window.

[0067] These eligible records are temporarily marked by the system as "unscanned handover tags." Each unscanned handover tag carries a complete historical traceability chain for the waste package before its identity was lost, including at least: a globally unique ID, the legally defined category of medical waste (such as "infectious"), the weight at the previous node, the handover time, and the issuing department. In essence, it is a task to be verified where "the electronic identity is clear, but the physical entity has not appeared as scheduled according to the digital instructions."

[0068] Building global matching pairs: Next, the system enters a key logical construction step: it performs a full connection combination of all loss labels generated on-site at this node and all unscanned handover labels obtained from the database query to form a set of "matching pairs" to be analyzed.

[0069] Specifically, if this node generates M loss labels and there are N unscanned handover labels in the historical record, the system will construct M×N candidate matching pairs.

[0070] For example, at the "Medical Waste Temporary Storage Room on the 5th Floor of the Inpatient Department" node, the operator found two bags of waste with QR codes that could not be read.

[0071] The system generates a loss label L1 for the first bag (collected information: weight 5.2kg, image features: "yellow polyethylene plastic bag with traces of liquid seepage").

[0072] Generate a loss label L2 for the second bag (collected information: weight 3.0kg, image features are "disposable rigid plastic sharps box, red lid").

[0073] Meanwhile, the system query records from the previous nodes "Orthopedic Ward" and "Operating Room" revealed three bags of registered waste that should have arrived at this temporary storage room during this transfer but were not scanned:

[0074] Unscanned handover label U1 (historical information: ID "001", category "infectious waste", weighed 5.0 kg in the ward).

[0075] Unscanned handover label U2 (historical information: ID "008", category "damaging waste", weighed 3.1 kg in the operating room).

[0076] Unscanned handover label U3 (historical information: ID "015", category "infectious waste", weighed 4.8 kg in the ward).

[0077] The system then automatically generated all six matching pairs to be processed: (L1, U1), (L1, U2), (L1, U3), (L2, U1), (L2, U2), (L2, U3).

[0078] S102. Test the pre-trained neural network using all known labels of this node, and determine the feature weights of the corresponding features based on the classification accuracy of each pre-trained neural network.

[0079] In this embodiment, the pre-trained neural network is tested using all known labels of this node, and the feature weights of the corresponding features are determined based on the classification accuracy of each pre-trained neural network. Specifically, this includes:

[0080] Input the packaged images of all known labels of this node into each pre-trained neural network to obtain the classification results of each pre-trained neural network for each known label;

[0081] The classification result of each known label is compared with the true category of that known label;

[0082] Count the number of known labels correctly classified by each pre-trained neural network;

[0083] Based on the number of correctly classified known labels, feature weights are assigned to the features corresponding to each pre-trained neural network. The assigned feature weights are positively correlated with the number of correctly classified labels.

[0084] Based on the number of correctly classified known labels, feature weights are assigned to the features corresponding to each pre-trained neural network, specifically including:

[0085] The classification accuracy of a network is obtained by dividing the number of known labels correctly classified by each pre-trained neural network by the total number of known labels of that node, and this classification accuracy is used as the feature weight of its corresponding feature.

[0086] For example, after forming a matching pair, all medical waste packages that can be scanned and identified normally in the current node are first processed. These medical waste packages are marked with "known labels," meaning their packaging images, weights, and manually verified waste category information are all known and reliable. Multiple pre-trained neural networks are specialized models pre-trained on a large-scale medical waste image dataset. Each network is designed to analyze waste packaging features from different visual dimensions, and its training objective is to accurately identify the five legal categories of medical waste (infectious, pathological, traumatic, pharmaceutical, and chemical). The specialization of these networks lies in their specialized division of labor. Each network performs deep feature learning for specific visual dimensions of medical waste packaging. Before deployment, these networks have been pre-trained on a database of millions of medical waste images covering various lighting conditions, shooting angles, and packaging conditions.

[0087] The core of the network weight dynamic calibration mechanism lies in using all known labels of the current node (i.e., successfully scanned but unusable packets) as a "standard answer set" to test the three networks in real time. Specifically, the true category of each known label is compared with the network's predicted category, and the number of correct classifications for each network is counted. This number directly reflects which visual feature dimension has the highest discriminative reliability under the current environmental conditions (such as specific lighting, shooting angle, background interference, etc.).

[0088] The following steps are taken to perform on-site performance evaluation and weight calibration of the pre-trained neural network using this set of known data:

[0089] First, the packaging images of all known labels for the current node are sequentially input into pre-trained independent neural networks (e.g., three pre-trained neural networks). These three networks focus on extracting visual features in different dimensions: the first network (Net-M) mainly analyzes the packaging material and shape (e.g., determining whether it is a soft plastic bag or a hard sharps container); the second network (Net-C) focuses on recognizing packaging colors and markings (e.g., recognizing the yellow of infectious waste or the red of sharps waste); and the third network (Net-A) is used to detect auxiliary features (such as specific text labels, biohazard symbols, or sealing methods).

[0090] Each neural network outputs a waste category prediction for each input image. Then, the prediction of each network is automatically compared with the true category of the image's known label (i.e., the standard category confirmed and entered manually). For example, if Net-M predicts the category of 100 images with known labels, and 85 of them match the true category, then the number of correctly classified images is recorded as 85.

[0091] The number of known labels correctly classified by each of the three neural networks is counted. Then, dynamic feature weights are assigned to the features extracted by each network based on this number.

[0092] The calculation method for feature weights (for the i-th neural network (corresponding to the i-th type of visual feature, such as material, color, or auxiliary features), its feature weights are... The calculation formula can be:

[0093] ;

[0094] in,

[0095] ;

[0096] In the formula, Let represent the feature weight corresponding to the i-th feature class, and 'a' represent the a-th known label. This represents the consistency discriminant function. This indicates that after the a-th known label is input into the i-th neural network, its output category matches the true category of that label. This indicates that when the a-th known label is input into the i-th neural network, its output category is inconsistent with the true category of the label.

[0097] The actual calculation process of this formula is as follows: For each known label of the current node, its packaged image is sequentially input into the i-th neural network, and the network's output category is compared with the true category registered in the system for that label. If they match, the count is incremented by 1; otherwise, it is incremented by 0. After traversing all known labels, the accumulated sum is the feature weight corresponding to the feature of the neural network.

[0098] S103. Based on the feature weight, perform cluster analysis on all loss labels to obtain the predicted category of each loss label.

[0099] In this embodiment, cluster analysis is performed on all loss labels based on the feature weights to obtain the predicted category of each loss label, specifically including:

[0100] For any two loss labels, input the packaged images of these two loss labels into each pre-trained neural network respectively;

[0101] For each pre-trained neural network, determine whether its classification results for the two loss labels are the same;

[0102] If they are the same, the feature weights corresponding to the pre-trained neural network are normalized and included in the clustering contribution value of the two loss labels; if they are different, zero is included in the clustering contribution value.

[0103] The clustering contribution values ​​of the two loss labels are summed to obtain the similarity score between the two loss labels;

[0104] Based on the similarity scores between all pairs of loss labels, clustering algorithms are used to cluster all loss labels to obtain classification clusters;

[0105] Each classification cluster is matched with each predefined standard category of medical waste to determine the medical waste category corresponding to each classification cluster, and this medical waste category is determined as the predicted category for each loss label in that classification cluster.

[0106] The clustering algorithm used in this embodiment is the K-means clustering algorithm, and the clustering parameter of the clustering algorithm is set to 5.

[0107] For example, after obtaining the feature weights corresponding to each visual feature, the visual similarity between all loss labels is quantified pairwise, and automatic clustering is performed based on this, thereby assigning a predicted category to each loss label. The specific steps are as follows:

[0108] For any two loss labels, their visual similarity is calculated by weighted ensemble of the discrimination results from each neural network. The specific process is as follows:

[0109] The wrapped images with the two lost labels are input into each pre-trained neural network, and the output predicted category of each network is recorded. The formula for calculating the similarity score between the two labels can be:

[0110] ;

[0111] in, Let represent the similarity score between the g-th loss label and the f-th loss label, and let i represent the i-th pre-trained neural network. This represents the feature weight corresponding to the i-th type of feature. Let f represent the consistency indicator function between the g-th loss label and the f-th loss label in the i-th neural network. This function is used when the i-th neural network predicts the same category for the g-th and f-th loss labels. It is 1 if it is true, otherwise it is 0.

[0112] Optionally, in this embodiment, the feature weights corresponding to the pre-trained neural network are normalized by using... The normalization method can be replaced, and there are no restrictions here.

[0113] The similarity score is calculated using a weighted voting mechanism, where the voting weight of each network is determined by its feature weights. After calculating the pairwise similarity scores between all loss labels, a complete similarity matrix is ​​formed.

[0114] Considering that medical waste is classified into five legal categories—infectious, pathological, traumatic, pharmaceutical, and chemical—the number of clusters is preset to K=5, and the K-means algorithm is used for clustering.

[0115] First, convert the similarity matrix into a distance matrix. The conversion rule can be: distance = 1 - similarity score.

[0116] Using the feature vector extracted by the neural network for each loss label as a sample, five cluster centers are randomly initialized.

[0117] Based on the distance matrix, sample allocation and center update are performed iteratively until the center points are stable, and finally 5 preliminary clusters are output.

[0118] To avoid missing categories or oversegmentation caused by forced clustering, perform the following matching and merging operations:

[0119] Calculate the mean of the feature vectors of all samples within each initial cluster to obtain the cluster center feature vector.

[0120] The pre-built standard optimal image library, which contains standard images and feature vectors of five types of medical waste, is used as a standard template.

[0121] Using the visual similarity calculation method described above, the similarity score between each cluster center feature vector and each standard template is calculated.

[0122] For each cluster, the medical waste category corresponding to the standard template with the highest similarity score is assigned as the matching category for that cluster.

[0123] Iterate through all clusters, merge clusters with the same matching category, and integrate their sample and feature data to form a unified final cluster. After merging, each final cluster corresponds to a unique medical waste category, which serves as the predicted category for all loss labels within the cluster.

[0124] S104. For each matching pair, compare the predicted category of the loss label with the historical category of the unscanned handover label in the matching pair.

[0125] For example, after completing the cluster analysis of all loss labels and obtaining their respective predicted categories, a preliminary category-level screening is required for each matching pair.

[0126] For each matching pair consisting of the loss label and the unscanned handover label, perform the following operation:

[0127] Extract the predicted category of the lost label (i.e., the medical waste category assigned to it through the aforementioned clustering analysis steps, such as "infectious waste"). At the same time, extract the historical category of the unscanned handover label (i.e., the actual category recorded by the label at its most recent successfully scanned handover node, which is also one of the standard categories such as "infectious waste" or "damaging waste").

[0128] The two categories are directly compared to determine if they are the same. The comparison result will produce a binary class consistency judgment:

[0129] If they are the same: the matching pair is determined to be "of the same category". This means that from the perspective of waste classification, the lost tag and the unscanned handover tag belong to the same category, and there is a possibility that they are the same physical entity. This matching pair will be retained and will enter the subsequent matching cost calculation process.

[0130] If they are different, the matching pair is determined to be "incompatible in category". This indicates that the two do not match in basic classification and belong to different types of medical waste, so they are highly unlikely to be the same item. For such matching pairs, a very large cost representing "unmatched" will be directly assigned to them, and all subsequent complex timing and weight calculations will be skipped, thereby significantly improving the system's processing efficiency.

[0131] S105. If consistent, then using all known labels of this node as a reference, calculate the current weight difference between each known label and the lost label, and the historical weight difference between each known label and the unscanned overlapping label. Calculate the preliminary matching metric based on the consistency of the current weight difference and the historical weight difference. Obtain the weight stability weight according to the predicted category of the lost label, and weight the absolute weight difference between the two labels in the matching pair to generate a weight penalty term. Combine the preliminary matching metric and the weight penalty term to generate the final matching cost for each matching pair. If inconsistent, generate a maximum cost for each matching pair as the final matching cost.

[0132] In this embodiment, using all known tags of this node as references, the current weight difference between each known tag and the lost tag, and the historical weight difference between each known tag and the unscanned overlapping tag are calculated. Based on the consistency of the current weight difference and the historical weight difference, a preliminary matching metric is calculated, specifically including:

[0133] For each known tag in this node, calculate the first weight difference between its current weight and the current weight of the lost tag, and calculate the second weight difference between its historical weight in the previous node and the historical weight of the unscanned handover tag.

[0134] Calculate the absolute value of the difference between the first weight difference and the second weight difference;

[0135] The matching contribution of the known label to the matching pair is determined based on the magnitude of the absolute value, where the absolute value and the matching contribution are negatively correlated.

[0136] Calculate the absolute value of the difference between the current weight of the known tag and the current weight of the lost tag;

[0137] A weight is assigned to the known label based on the absolute value of the weight difference, wherein the absolute value of the weight difference and the assigned weight are negatively correlated.

[0138] The matching contribution value of the known label is weighted according to the assigned weights to obtain the weighted contribution of the known label;

[0139] Sum the weighted contributions of all known labels and use the summation result as a preliminary matching measure.

[0140] The weight stability weight is obtained based on the predicted category of the loss label. The absolute weight difference between the two labels in the matching pair is weighted to generate a weight penalty term, which specifically includes:

[0141] Based on the predicted category of the loss label, the corresponding weight stability weight is obtained by querying the preset weight stability weight mapping table. The preset weight stability weight mapping table is established based on historical data analysis, and the weight stability weight in the preset weight stability weight mapping table is positively correlated with the stability of weight change of medical waste category.

[0142] Calculate the absolute value of the difference between the current weight of the lost tag in the matching pair and the historical weight of the unscanned handover tag, and normalize it to obtain the absolute weight difference.

[0143] Multiply the absolute weight difference by the weight stability weight to obtain the weight penalty term.

[0144] The final matching cost for each matching pair is generated by combining the initial matching metric with a weight penalty term, specifically including:

[0145] The initial matching metric is converted into an initial matching cost, where the initial matching metric and the initial matching cost are negatively correlated.

[0146] Multiply the initial matching cost by the weight penalty term to obtain the final matching cost of the matching pair.

[0147] For example, for each matching pair determined to be of the same category (i.e., the predicted category of the lost label is the same as the historical category of the unscanned handover label), its comprehensive final matching cost needs to be calculated. This cost integrates information from two dimensions: the flow time series pattern and the physical weight characteristics.

[0148] Optionally, in a preferred embodiment, the negative correlation is represented by the exp function in the following formula. In some other feasible embodiments, other negative correlation functions can be used instead, such as inverse proportional functions, Gaussian kernel functions, etc. The appropriate function form can be selected according to the actual data distribution, computational efficiency requirements, or system optimization goals, and there are no restrictions here.

[0149] In a standardized medical waste management system, known labels represent standard samples that have undergone complete barcode scanning and transfer with accurate information records. Their complete transfer records (including timestamps, weight, and location) provide a reliable reference for analyzing the transfer patterns of batches of waste.

[0150] When evaluating a matching pair consisting of a lost tag (a waste package that the current node cannot scan) and an unscanned handover tag (a waste package registered at the previous node but not present at this node), the temporal matching between the two can be indirectly inferred by using known tags as a "bridge." The core logic is that medical waste generated in the same batch should have highly similar time intervals and weight change patterns during its transfer from the previous node to the current node.

[0151] Specifically, a known tag is selected, and its weight difference with the lost tag at the current node, as well as its historical weight difference with the unscanned intersection tag at the previous node, are calculated. If the lost tag and the unscanned intersection tag are actually the same item, then the difference between these two weight differences should be minimal, because the weight change pattern of the known tag they both reference is fixed. In other words, the "weight change rhythm from the previous node to the current node" embodied by the known tag should be completely synchronized with the weight relationship change rhythm of the two tags in the matching pair relative to the known tag.

[0152] If calculations reveal that multiple known tags indicate this high degree of synchronicity, it means that the flow rhythm of lost tags and unscanned handover tags closely matches the overall pattern of the current batch of waste, indicating a very high probability that they belong to the same item, i.e., high temporal matching. Conversely, if most known tags indicate poor synchronicity, it suggests a significant deviation in their flow sequence, likely belonging to batches processed at different times, resulting in low temporal matching.

[0153] For the loss label and unscanned handover label in a matching pair, and a known label, calculate the matching contribution of the known label to the current matching pair. The calculation method can be as follows:

[0154] ;

[0155] in, Let exp represent the matching contribution of the a-th known label to each matching pair consisting of the loss label and each unscanned intersection label. exp represents an exponential function with base e (which can be used here to handle the negative correlation between absolute value and matching contribution value, or can be replaced by other inverse proportional functions). This represents the weight of the a-th known label at the current node. This represents the weight of the v-th loss label at the current node. This represents the weight of the a-th known label at the previous node. This represents the weighing weight of the u-th unscanned handover tag at the previous node.

[0156] The closer this value is to 1, the more it indicates that the known weight change pattern reflected by the tags supports the idea that the v-th lost tag and the u-th unscanned handover tag are the same item.

[0157] Not all known labels contribute equally. Known labels whose weight is closer to the current weight of the v-th loss label are more likely to belong to the same batch, and their contributions should be given higher weight. Therefore, a weight should be assigned to the a-th known label. The calculation method can be as follows:

[0158] ;

[0159] This represents the weight assigned to the a-th known label based on the absolute value of the difference between the weighing weight of the a-th known label at the current node and the weighing weight of the v-th lost label at the current node.

[0160] By combining the weighted contributions of all known labels, a preliminary matching measure of the matching pair is obtained. The calculation formula can be:

[0161] ;

[0162] The larger the initial matching metric value (maximum 1), the higher the temporal matching.

[0163] Different categories of medical waste exhibit significant differences in weight stability due to variations in packaging characteristics, contents, and storage conditions. For example, infectious waste (such as cotton swabs and dressings) is typically packaged in soft plastic bags, and its weight may experience acceptable slight fluctuations during transport due to liquid evaporation, minor leakage, or environmental adsorption. In contrast, sharps waste (such as sharps containers) uses rigid, sealed containers with fixed and non-volatile contents, resulting in a weight that remains almost constant in the short term, with theoretically minimal allowable deviation. Other categories, such as pharmaceutical and chemical waste, also possess their own unique weight variation characteristics.

[0164] Applying a uniform weight difference tolerance threshold to all waste categories for matching will inevitably lead to misjudgments: infectious waste may be misjudged as mismatched due to minor weight changes caused by normal evaporation; and for damaged waste, a lenient tolerance threshold may overlook the actual weight loss risk caused by packaging damage. Therefore, it is necessary to set differentiated weight stability weights for each type of waste based on its physical characteristics to achieve more accurate matching.

[0165] When determining the weight stability weights for each category, the following principles are followed: Based on historical handover data, analyze the weight changes between adjacent handover nodes for all known tag-matching pairs (i.e., waste that has successfully completed barcode scanning and handover) of a certain category. If the average weight change of this type of waste is small, it indicates high weight stability, and a larger weight stability weight should be assigned, meaning a stronger penalty for weight differences in the matching calculation, i.e., requiring stricter weight consistency. Conversely, if the average weight change is large, it indicates strong weight volatility, and a smaller weight stability weight should be assigned, allowing for greater tolerance of weight differences.

[0166] Furthermore, when analyzing historical weight change data, outlier samples must be excluded to ensure the reliability of weight calculation. Specifically, for all known label matching pairs of a certain category, a sample is only included in the valid data for calculating the average weight change if its weight change between adjacent nodes does not exceed 5% of the minimum weight change among samples in that category. If the weight change of a sample exceeds this threshold, it is considered an abnormal fluctuation (possibly due to recording errors or special events) and is excluded. The average weight change calculated based on the filtered valid samples is then normalized and finally mapped to a weight stability weight that reflects the weight stability of that category.

[0167] Optionally, the weight stability weight reflecting this type of weight stability can also be established based on other characteristics of historical data. For example, the variance or standard deviation of the weight change of a certain type of medical waste between adjacent nodes can be calculated. If the variance is small, it indicates that the weight stability of this type of waste is high, and a larger weight stability weight should be assigned. If the range of the weight change of a certain type of waste is small, it indicates that its weight fluctuation range is limited and its weight stability is high, etc. This embodiment only provides a preferred embodiment, that is, the weight stability weight mapping table can be determined by availability criteria. Generated based on historical data.

[0168] Based on the above analysis and data filtering rules, the weight stability weight of each medical waste category can be calculated using the following quantitative method. :

[0169] ;

[0170] in,

[0171] ;

[0172] In the formula, This represents the weight stability weight of the Mth medical waste category. This represents the maximum and minimum value normalization function. Let represent the availability criteria for the a-th known label within the M-th category of medical waste. This represents the weighing weight of the a-th known label within the M-th medical waste category in the current process. This represents the weighing weight of the a-th known label in the M-th medical waste category in the previous process. This represents the minimum value function.

[0173] As shown above, the weight stability weight for each category of medical waste can be determined.

[0174] After obtaining the initial matching cost (its temporal matching degree), physical weight information needs to be further incorporated for more refined screening. The core idea is that if two tags represent the same item, their weights should be highly similar. Therefore, introducing weight difference as a penalty factor can correct for potential biases arising from relying solely on temporal information. Specifically, even if a matching pair is highly temporally matched (with a small initial matching cost), a significant weight difference should reduce its matching reliability; conversely, if the weights are almost identical, even a moderate temporal matching degree may increase its matching probability.

[0175] Therefore, weight penalty item The calculation formula is:

[0176] ;

[0177] in, This represents the maximum difference between the weighing weight of all lost tags at the current node and the weighing weight of all unscanned tags at the previous node. This represents the weight stability weight of the medical waste category to which the predicted category of the v-th loss label belongs.

[0178] The final matching cost of the v-th loss label and the u-th unscanned intersection label in the matching pair The calculation method can be:

[0179] ;

[0180] This formula achieves an organic coupling of temporal and weight information: the initial matching cost reflects the degree of matching in the flow rhythm, while the weight penalty term introduces a correction based on physical characteristics. The final matching cost, constituted by both, provides a more comprehensive and reliable quantitative basis for subsequent global optimal matching using algorithms such as the Hungarian algorithm.

[0181] S106. Construct a cost matrix based on the final matching cost of all matching pairs, use the Hungarian algorithm to perform global optimal matching, and determine the full-process traceability information for the loss label of successful matching.

[0182] In this embodiment, a cost matrix is ​​constructed based on the final matching costs of all matching pairs. The Hungarian algorithm is used for globally optimal matching, and full-process traceability information is determined for the loss labels of successful matches. Specifically, this includes:

[0183] The matrix is ​​composed of rows of all loss labels and columns of all unscanned intersection labels.

[0184] The final matching cost of each matching pair is filled into the matrix element position corresponding to the intersection of the row of the loss label and the column of the unscanned intersection label in that matching pair, to obtain the initial cost matrix;

[0185] If the number of lost labels is not equal to the number of unscanned handover labels, then add rows or columns filled with preset values ​​to the matrix to expand the initial cost matrix into a square matrix, thus obtaining the cost matrix.

[0186] The cost matrix is ​​input into the Hungarian algorithm for calculation, and the globally optimal matching scheme with the minimum total cost is obtained.

[0187] Based on the optimal matching scheme, determine the unscanned handover label that matches each lost label, and perform data association for each successfully matched pair of labels;

[0188] The complete historical traceability information of the unscanned handover label that was successfully matched from the node to the previous node is assigned to the matching loss label, and the information of the loss label of this node is completed to obtain the full-process traceability information of the loss label in this node.

[0189] For loss labels that fail to match successfully in the optimal matching scheme or for unscanned handover labels, generate a manual verification warning.

[0190] For example, after calculating the final matching cost for all matching pairs, the process proceeds to the global decision-making and data repair phase. The core objective of this phase is to find an optimal matching scheme with the minimum total cost from all possible matching combinations, and to repair the interrupted tracing chain accordingly.

[0191] Construct a cost matrix with all loss labels as rows and all unscanned intersection labels as columns. The element in the i-th row and j-th column of the matrix represents the final matching cost between the v-th loss label and the u-th unscanned intersection label.

[0192] For any matching pair, if the historical categories of the v-th loss label and the u-th unscanned intersection label are inconsistent, the value of the matrix element is directly set to a maximum value (such as positive infinity) representing "unmatchable", thereby eliminating such obviously unreasonable combinations in advance at the algorithm level.

[0193] If the number of lost labels m is not equal to the number of unscanned intersection labels n, the cost matrix is ​​expanded into a max(m, n) square matrix by adding rows or columns filled with specific values ​​(such as 0) to meet the input requirements of the Hungarian algorithm. The virtual labels corresponding to the newly added rows and columns will be marked as "awaiting manual verification".

[0194] The preprocessed cost matrix is ​​input into the Hungarian algorithm. The algorithm iteratively calculates through a series of standard steps (such as row reduction, column reduction, finding the minimum number of lines covering all zero elements, adjusting the matrix, finding augmenting paths, etc.) and finally solves for an optimal 0-1 assignment matrix.

[0195] In this result matrix, elements with a value of 1 represent the optimal matching pair recommended by the algorithm. For example, if the element in the i-th row and j-th column of the result matrix is ​​1, it indicates that the algorithm determines that the v-th loss label should be matched with the u-th unscanned intersection label. This matching scheme guarantees that the sum of the final matching costs for all successfully matched label pairs is minimized.

[0196] Based on the optimal matching scheme output by the Hungarian algorithm, the system performs automatic repair and early warning: for matching pairs where the algorithm successfully matches and the final matching cost is lower than a preset confidence threshold, the system performs data association operations. Specifically, the complete historical traceability information (including the originating department, generation time, and previous handover records) carried by the u-th unscanned handover tag is assigned to the corresponding v-th lost tag. Subsequently, this association is synchronously updated to the full-process traceability system database, thereby completing the data chain interrupted by tag damage and forming a continuous and complete electronic traceability record.

[0197] Manual verification alert: The system will trigger a manual verification alert in the following two situations:

[0198] The algorithm found a match, but the final matching cost exceeded the preset threshold. This indicates that although the algorithm provided a matching suggestion, its confidence level was low, and there may be doubts, requiring final human confirmation.

[0199] Loss labels that were not successfully matched by the algorithm or unscanned handover labels (including virtual labels generated due to matrix completion) are considered "unsolved mysteries" and require manual intervention to investigate the cause, such as checking whether they are newly generated waste, whether the labels were misread, or whether there are other management loopholes.

[0200] In summary, this invention, through multi-dimensional intelligent fusion and global optimization matching, achieves automatic repair and risk warning of the traceability data chain in the fault scenario of damaged physical tags on medical waste. Its innovation lies in: first, dynamically calibrating visual feature weights using known tags on-site, enabling the system to adapt to environmental changes and robustly complete image classification of lost tags; second, uniquely using known tags as a "temporal bridge," indirectly evaluating matching by analyzing the consistency of weight change patterns, overcoming the limitations of directly relying on timestamps; and finally, introducing weight stability weights based on category physical characteristics, achieving refined matching "tailored to specific categories." This method ultimately solves the global optimal matching problem through the Hungarian algorithm and couples automatic repair and manual verification mechanisms, fundamentally solving the problem of information chain breakage caused by tag damage, significantly improving the data integrity, management reliability, and timeliness of risk warning in the entire traceability system, while greatly reducing the cost of manual verification.

[0201] The present invention also proposes a traceability and risk warning system for the entire process of medical waste. Please refer to Figure 2, which shows a structural diagram of a traceability and risk warning system for the entire process of medical waste provided by an embodiment of the present invention. The system includes: a pretreatment module 101, a data analysis module 102, and a matching module 103.

[0202] Preprocessing module 101 is used to generate loss tags for medical waste packages that cannot be scanned at the handover node, and form a matching pair with the unscanned handover tags that have been recorded at the previous node and have not appeared at this node;

[0203] The data analysis module 102 is used to test the pre-trained neural network using all known labels of this node, and to determine the feature weight of the corresponding feature based on the classification accuracy of each pre-trained neural network.

[0204] Based on this feature weight, cluster analysis is performed on all loss labels to obtain the predicted category of each loss label;

[0205] For each matching pair, compare the predicted category of the loss label with the historical category of the unscanned handover label in that matching pair;

[0206] If consistent, then using all known labels of this node as references, calculate the current weight difference between each known label and the lost label, and the historical weight difference between each known label and the unscanned overlapping label. Calculate the preliminary matching metric based on the consistency of the current weight difference and the historical weight difference. Obtain the weight stability weight according to the predicted category of the lost label, and weight the absolute weight difference between the two labels in the matching pair to generate a weight penalty term. Combine the preliminary matching metric and the weight penalty term to generate the final matching cost for each matching pair.

[0207] If they are inconsistent, a maximum cost is generated for each matching pair as the final matching cost;

[0208] The matching module 103 is used to construct a cost matrix based on the final matching cost of all matching pairs, use the Hungarian algorithm to perform global optimal matching, and determine the full-process traceability information for the loss label of successful matching.

[0209] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the traceability and risk warning system for the entire process of medical waste and the method embodiment for traceability and risk warning for the entire process of medical waste provided in the above embodiments belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0210] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0211] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0212] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for traceability and risk warning throughout the entire process of medical waste disposal, characterized in that, include: At the handover node, a loss label is generated for medical waste packages that cannot be scanned, and a matching pair is formed with the unscanned handover label that has been recorded in the previous node but has not appeared in this node; the pre-trained neural network is tested using all known labels of this node, and the feature weight of its corresponding feature is determined based on the classification accuracy of each pre-trained neural network. Based on this feature weight, cluster analysis is performed on all loss labels to obtain the predicted category for each loss label. Specifically, this includes: for any two loss labels, inputting the packaging images of these two loss labels into each pre-trained neural network; for each pre-trained neural network, determining whether its classification results for the two loss labels are the same; if they are the same, normalizing the feature weights corresponding to the pre-trained neural network and including them in the clustering contribution value of the two loss labels; if they are different, zero is included in the clustering contribution value; summing the clustering contribution values ​​of the two loss labels to obtain the similarity score between the two loss labels; and based on the pairwise similarity scores of all loss labels... Clustering algorithms are used to cluster all lost labels to obtain classification clusters. Each classification cluster is matched with each predefined medical waste standard category to determine the corresponding medical waste category. This medical waste category is then used as the predicted category for each lost label within that cluster. For each matching pair, the predicted category of the lost label is compared with the historical category of the unscanned handover label in that pair. If they match, all known labels of this node are used as references to calculate the current weight difference between each known label and the lost label, as well as the historical weight difference between each known label and the unscanned handover label. Based on the consistency between the current weight difference and the historical weight difference, a preliminary matching metric is calculated. The process includes: for each known label in the current node, calculating a first weight difference between its current weight and the current weight of the lost label, and calculating a second weight difference between its historical weight in the previous node and the historical weight of the unscanned handover label; calculating the absolute value of the difference between the first and second weight differences; determining the matching contribution value of the known label to the matching pair based on the magnitude of the absolute value, wherein the absolute value and the matching contribution value are negatively correlated; calculating the absolute value of the difference between the current weight of the known label and the current weight of the lost label; assigning a weight to the known label based on the magnitude of the absolute value of the weight difference, wherein the absolute value of the weight difference and the assigned weight are negatively correlated; and using the assigned weight as the basis for further calculations. The weighted contribution of the known label is weighted by the assigned weights to obtain the weighted contribution of the known label. The weighted contributions of all known labels are summed, and the sum is determined as the preliminary matching measure. The weight stability weight is obtained according to the predicted category of the loss label, and the absolute weight difference between the two labels in the matching pair is weighted to generate a weight penalty term. The preliminary matching measure and the weight penalty term are fused to generate the final matching cost of each matching pair. If they are inconsistent, a maximum cost is generated for each matching pair as the final matching cost. A cost matrix is ​​constructed based on the final matching costs of all matching pairs, and the Hungarian algorithm is used to perform global optimal matching to determine the full-process traceability information for the successfully matched loss label.

2. The method for traceability and risk warning of medical waste throughout the entire process according to claim 1, characterized in that, The process of testing pre-trained neural networks using all known labels of the current node and determining the feature weights of corresponding features based on the classification accuracy of each pre-trained neural network includes: inputting the packaged images of all known labels of the current node into each pre-trained neural network to obtain the classification results of each pre-trained neural network for each known label; comparing the classification results of each known label with the true category of the known label; counting the number of known labels correctly classified by each pre-trained neural network; and assigning feature weights to the features corresponding to each pre-trained neural network based on the number of correctly classified known labels, wherein the assigned feature weights are positively correlated with the number of correctly classified features.

3. The method for traceability and risk warning of medical waste throughout the entire process according to claim 1, characterized in that, The step of obtaining a weight stability weight based on the predicted category of the loss label and weighting the absolute weight difference between the two labels in the matching pair to generate a weight penalty term specifically includes: querying a preset weight stability weight mapping table based on the predicted category of the loss label to obtain the corresponding weight stability weight, wherein the preset weight stability weight mapping table is established based on historical data analysis, and the weight stability weight in the preset weight stability weight mapping table is positively correlated with the stability of weight change of medical waste category; calculating the absolute value of the difference between the current weight of the loss label in the matching pair and the historical weight of the unscanned handover label, and performing normalization processing to obtain the absolute weight difference; multiplying the absolute weight difference by the weight stability weight to obtain the weight penalty term.

4. The method for traceability and risk warning of medical waste throughout the entire process according to claim 1, characterized in that, The process of generating the final matching cost for each matching pair by fusing the initial matching metric and the weight penalty term specifically includes: converting the initial matching metric into an initial matching cost, wherein the initial matching metric and the initial matching cost are negatively correlated; and multiplying the initial matching cost and the weight penalty term to obtain the final matching cost for the matching pair.

5. The method for traceability and risk warning of medical waste throughout the entire process according to claim 1, characterized in that, The process involves constructing a cost matrix based on the final matching costs of all matching pairs, using the Hungarian algorithm for globally optimal matching, and determining full-process traceability information for successfully matched loss labels. Specifically, this includes: using all loss labels as rows and all unscanned intersection labels as columns; filling the matrix with the final matching cost of each matching pair into the matrix element corresponding to the intersection of the row containing the loss label and the column containing the unscanned intersection label, thus obtaining the initial cost matrix; if the number of loss labels is not equal to the number of unscanned intersection labels, adding rows or columns filled with preset values ​​to the matrix, expanding the initial cost matrix into a square matrix, and obtaining the final cost. The cost matrix is ​​input into the Hungarian algorithm for calculation to obtain the globally optimal matching scheme with the minimum total cost. Based on the optimal matching scheme, the unscanned handover labels matched by each loss label are determined, and data association is performed for each successfully matched pair of labels. The complete historical traceability information from the generation node to the previous node carried by the successfully matched unscanned handover labels is assigned to the matching loss labels, and the information of the loss labels in this node is completed to obtain the full-process traceability information of the loss labels in this node. For loss labels or unscanned handover labels that fail to match successfully in the optimal matching scheme, a manual verification warning is generated.

6. The method for traceability and risk warning of medical waste throughout the entire process according to claim 2, characterized in that, The method of assigning feature weights to the features corresponding to each pre-trained neural network based on the number of correctly classified known labels includes: dividing the number of correctly classified known labels of each pre-trained neural network by the total number of known labels of the current node to obtain the classification accuracy of the network, and using the classification accuracy as the feature weight of its corresponding feature.

7. The method for traceability and risk warning of medical waste throughout the entire process according to claim 1, characterized in that, The clustering algorithm is the K-means clustering algorithm, and the clustering parameter of the clustering algorithm is set to 5.

8. A traceability and risk warning system for the entire medical waste process, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for tracing and risk warning of the entire medical waste process as described in any one of claims 1-7.

Images