Method, intelligent agent and system for supervising compliance of hazardous waste label

Abstract

The invention provides a method, an intelligent agent and a system for supervising the compliance of a hazardous waste label, and the method comprises the steps: analyzing the comprehensive information of the hazardous waste label from an obtained image of the hazardous waste label through a large model, and enabling the large model to be obtained through the training of the image of the hazardous waste label, analyzing license key information of the obtained pollution discharge license through a large model; acquiring a hazardous waste management plan and a management ledger; the large model examines the compliance of the comprehensive information of the hazardous waste label according to an input preset rule base, and performs cross examination on the consistency according to the obtained hazardous waste management plan and management ledger; and outputting an examination result and a rectification opinion. The technical problems that in the prior art, time and labor are wasted, comprehensive coverage cannot be achieved, timeliness is insufficient, and the requirement for monitoring personnel is high in a manual on-site monitoring mode are solved, the accuracy of compliance monitoring of hazardous waste labels can be guaranteed, and meanwhile the monitoring efficiency and the intelligent level are improved.

Description

Technical Field

[0001] This invention relates to the field of environmental protection supervision, and in particular to a method, intelligent agent, and system for monitoring the compliance of hazardous waste labeling. Background Technology

[0002] Currently, the following are some existing technologies for verifying the compliance of hazardous waste labeling content during the generation, storage, self-utilization / disposal, and transfer of hazardous waste:

[0003] On-site inspection: Inspectors go to the site to check whether the content of the hazardous waste labels in the generation, storage, self-utilization / disposal and transfer of hazardous waste is compliant, and to check whether the content recorded on the hazardous waste labels is consistent with the corresponding content of the enterprise's discharge permit, hazardous waste management plan and hazardous waste management ledger.

[0004] Enterprise self-assessment and supervision: Enterprises upload photos of hazardous waste labels from the generation, storage, self-utilization / disposal, and transfer stages to the regulatory platform and conduct compliance self-inspections, voluntarily ensuring standardized management of hazardous waste. Inspectors can also gain a clear understanding of the enterprise's management situation through the self-assessment results.

[0005] AI-enhanced mobile terminal + macro recognition technology: Equipped with a mobile terminal featuring a macro lens (supporting 1 cm focus) and a high-resolution sensor (≥48 megapixels), specifically designed for photographing tiny labels. AI algorithms are optimized to enhance image details through deep learning models. The terminal has a built-in pre-defined rule library for hazardous waste label compliance verification, automatically checking whether the label text and graphics meet standards. The processed data is compared in real-time with the enterprise's discharge permit, hazardous waste management plan, and management ledger on the business platform to generate a compliance report.

[0006] IoT device linkage + automatic verification of electronic ledgers: Smart weighing equipment and access control sensors are deployed at the hazardous waste generation, storage, and transfer stages. When hazardous waste is generated, the smart weighing equipment automatically records the weight and writes it to an RFID tag, simultaneously generating an electronic ledger. Transport vehicles are equipped with positioning devices and onboard weighing equipment, automatically linking tag data with the transport trajectory and vehicle load. Operating units confirm receipt by scanning a code, and the system automatically verifies the electronic ledger, ensuring consistency between the ledger and the goods. The monitoring platform automatically identifies anomalies (such as discrepancies between tag information and actual hazardous waste, or inconsistencies between transfer volume and ledger data) by comparing tag data, IoT-collected data, and enterprise-reported data.

[0007] The problems with the above-mentioned existing technologies include: Time-consuming and labor-intensive: Traditional problem and hidden danger inspections mainly rely on on-site inspections, requiring inspectors to visit various enterprises in person, which consumes a lot of time and manpower. Moreover, it is difficult to achieve high-frequency supervision when the inspection force is limited. There are many hazardous waste labels, and it is a huge workload for personnel to check them one by one. Moreover, most of the work is repetitive, and the efficiency of on-site inspections by personnel is low.

[0008] Inability to achieve full coverage: Enterprises that generate hazardous waste in my country are widely distributed, and traditional on-site inspections cannot cover them all at the same time; for enterprises with a large number of hazardous waste labels, on-site inspections by personnel are difficult to achieve full coverage, and only a few labels can be selected for spot checks.

[0009] Insufficient timeliness: On-site inspections can only be conducted at specific times (usually during working hours) and depend on the inspectors' work schedules, making it impossible to monitor hazardous waste labels in real time.

[0010] The requirements for inspectors are high: Hazardous waste labeling compliance inspection involves multiple inspection points and corresponding standards and regulations. Inspectors need to have a thorough understanding of these inspection points in order to ensure the accuracy of the inspection work.

[0011] On-site document review is inconvenient: Hazardous waste label compliance inspection involves verifying the label information against the company's discharge permit, hazardous waste management plan, and hazardous waste management ledger. Whether carrying paper copies of the discharge permit, hazardous waste management plan, and management ledger, or reviewing electronic versions of the above documents, it is inconvenient. Furthermore, some hazardous waste generation, storage, utilization, and disposal facilities do not allow mobile phones or other electronic devices to enter.

[0012] Automated photo acquisition is costly: Common methods like video surveillance and drone patrols cannot capture clear photos of hazardous waste labels, making it difficult for inspectors to accurately identify the text on the labels and compromising the accuracy of their inspections. Using macro lenses and high-resolution sensors to acquire high-definition photos is also expensive. Summary of the Invention

[0013] Based on the above problems, this invention proposes a method, intelligent agent, and system for monitoring the compliance of hazardous waste labels. It solves the technical problems of existing manual on-site monitoring methods, such as being time-consuming and labor-intensive, unable to provide comprehensive coverage, lacking timeliness, and requiring highly skilled inspectors. This invention can ensure the accuracy of compliance monitoring of hazardous waste labels, while improving monitoring efficiency and intelligence.

[0014] This invention proposes a method for monitoring compliance with hazardous waste labeling, comprising: The images of the acquired hazardous waste labels are parsed using a large model to extract comprehensive information about the hazardous waste labels. The large model is trained using images of hazardous waste labels. The comprehensive information about the hazardous waste labels includes at least: basic attribute information, hazardous characteristic warning information, management and responsibility information, and information traceability information. The management and responsibility information includes at least the name of the unit that generates or collects the hazardous waste and the date the hazardous waste is generated. Key information of the obtained discharge permits is extracted using a large model; Obtain the hazardous waste management plan and management ledger that correspond to the organization's name and the date the hazardous waste was generated; The large model reviews the compliance of comprehensive information on hazardous waste labels based on the input preset rule base, and cross-checks the consistency of key information on permits, hazardous waste management plans, management ledgers and comprehensive information on hazardous waste labels based on the obtained hazardous waste management plans and management ledgers. Output the review results and rectification suggestions.

[0015] Furthermore, the images of the acquired hazardous waste labels are parsed using a large model to extract comprehensive information about the hazardous waste labels. This large model is trained using images of hazardous waste labels and includes: The image is preprocessed to obtain a preprocessed image. The tensor of the preprocessed image is then input into the OCR visual model. The OCR visual model extracts features from the preprocessed image and generates multiple detection boxes with coordinates. Classify the content within each detection box and provide a confidence score; Filter out detection boxes with confidence scores lower than the preset confidence threshold; The duplicate detection boxes are removed using the nonmaximum suppression algorithm; For each deduplicated detection box, a structured data list is generated, consisting of the category, confidence score, border information, and category name.

[0016] In addition, after the structured data list is generated, the system checks whether a "hazardous characteristics" label exists in the structured data list. If it does not exist, the process is terminated, and a message is displayed indicating that the image of the hazardous waste label does not meet the requirements.

[0017] In addition, basic attribute information includes: name of hazardous waste, waste category and code, form of hazardous waste, main components and hazardous components, date of generation, weight and unit of hazardous waste; Hazard warning information must include at least a hazard icon and the words "Hazardous Characteristics"; Management and accountability information also includes the generating or collecting unit, contact person, and contact information; Information traceability includes hazardous waste digital identification codes and QR codes.

[0018] In addition, when parsing the hazard characteristic icons, the system checks for checkboxes. If a checkbox is found, a pairing relationship is established between the hazard characteristic icon and the adjacent checkbox, and the checked state of the checkbox is identified. Only the hazard characteristic icons corresponding to the checkboxes that are checked are identified as the actual hazard characteristics of the hazardous waste label.

[0019] In addition, key information about the obtained discharge permits was extracted using a large model, including: Combining a pre-set knowledge base, the large model pre-parses the discharge permit, locates the key page numbers in the discharge permit based on keywords in the pre-set knowledge base, converts the key page numbers into images and passes them to the multimodal model for parsing, and uses a pre-set rule base to calibrate the parsing results of the multimodal model. The parsing results include the permit validity period, unit name, unified social credit code, waste category and waste code.

[0020] In addition, a veto rule is set up. When the veto conditions are met, a preliminary verification report is generated for the hazardous waste label. If an image of a hazardous waste label is detected as a questionable label image, it is sent to an expert assistance platform. The expert assistance platform analyzes the questionable label image, marks the violations, uploads the evaluation evidence, and generates a final review report. The questionable label image with the marked violations, the evaluation evidence, and the final review report are used to train the large model.

[0021] The present invention also proposes an intelligent agent for monitoring the compliance of hazardous waste labeling, employing the method for monitoring the compliance of hazardous waste labeling as described in any of the preceding claims.

[0022] This invention also proposes a system for monitoring compliance with hazardous waste labeling, comprising: Camera module, weighing module, electronic data acquisition module, edge computing module, and cloud monitoring platform; The camera module, weighing module, and electronic data acquisition module are installed at the hazardous waste generation, storage, transfer, utilization, or disposal site. The weighing module weighs the hazardous waste and transmits the weight information to the electronic data acquisition module. The camera module acquires the image of the hazardous waste label and collects the time information, and then sends the image and time information to the data acquisition terminal. The data acquisition terminal transmits data to the edge computing module. The edge computing module uses a large model to parse the images of the acquired hazardous waste labels to obtain comprehensive information about the hazardous waste labels. The large model is trained using images of hazardous waste labels. The comprehensive information about hazardous waste labels includes at least: basic attribute information, hazardous characteristic warning information, management and responsibility information, and information traceability information. The management and responsibility information includes at least the name of the unit that generates or collects the hazardous waste and the date the hazardous waste is generated. Key information of the obtained discharge permits is extracted using a large model; The cloud-based monitoring platform obtains hazardous waste management plans and management ledgers that correspond to the unit's name and the date of hazardous waste generation. The large model reviews the compliance of comprehensive information on hazardous waste labels based on the input preset rule base, and cross-checks the consistency of key information on permits, hazardous waste management plans, management ledgers and comprehensive information on hazardous waste labels based on the obtained hazardous waste management plans and management ledgers. Output the review results and rectification suggestions.

[0023] This invention also proposes a system for monitoring compliance with hazardous waste labeling, comprising: A gateway and multiple microservice modules that communicate through the gateway; The microservice module should include at least: a tag parsing module, a license verification module, and a ledger retrieval module; The label parsing module receives images of hazardous waste labels sent from external sources and parses out the complete information of the hazardous waste labels, which is then sent to the ledger retrieval module. The complete information of the hazardous waste labels includes at least: basic attribute information, hazardous characteristic warning information, management and responsibility information, and information traceability information. The management and responsibility information includes at least the name of the unit that generates or collects the hazardous waste and the date the hazardous waste is generated. After the permit verification module obtains the discharge permit and parses out the key information of the permit, it sends it to the ledger retrieval module. The large model reviews the compliance of comprehensive information on hazardous waste labels based on the input preset rule base, and cross-checks the consistency of key information on permits, hazardous waste management plans, management ledgers and comprehensive information on hazardous waste labels based on the obtained hazardous waste management plans and management ledgers. Output the review results and rectification suggestions.

[0024] This invention solves the technical problems of existing manual on-site inspection methods, such as being time-consuming and labor-intensive, unable to provide comprehensive coverage, lacking timeliness, and requiring highly skilled inspectors. This invention can ensure the accuracy of compliance inspection of hazardous waste labels, while improving inspection efficiency and intelligence. Attached Figure Description

[0025] Figure 1 A flowchart illustrating a method for monitoring the compliance of hazardous waste labeling according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a hazardous characteristic icon without checkboxes provided in one embodiment of the present invention; Figure 3 A schematic diagram showing a checkbox on a hazardous characteristic icon provided in one embodiment of the present invention; Figure 4 The present invention provides a development process and business process for an intelligent agent for monitoring compliance with hazardous waste labeling, as an embodiment of the present invention. Detailed Implementation

[0026] The present invention will be further described in detail below with reference to specific embodiments and accompanying drawings. This description is intended only to illustrate specific embodiments of the invention and does not constitute any limitation on the invention. The scope of protection of the invention is defined by the claims.

[0027] Reference Figure 1 This invention proposes a method for monitoring compliance of hazardous waste labeling, comprising: Step S001: The images of the acquired hazardous waste labels are parsed using a large model to extract comprehensive information about the hazardous waste labels. The large model is trained using images of the hazardous waste labels. The comprehensive information about the hazardous waste labels includes at least: basic attribute information, hazardous characteristic warning information, management and responsibility information, and information traceability information. The management and responsibility information includes at least the name of the unit that generates or collects the hazardous waste and the date the hazardous waste is generated. Optionally, the management and responsibility information also includes: the unit's contact person and contact information.

[0028] Step S002: Extract key information from the obtained discharge permit using a large model. Step S003: Obtain the hazardous waste management plan and management ledger corresponding to the unit name and the date of hazardous waste generation; Step S004: The large model reviews the compliance of the comprehensive information of the hazardous waste label based on the input preset rule base, and cross-checks the consistency of the key information of the permit, the hazardous waste management plan, the management ledger and the comprehensive information of the hazardous waste label based on the obtained hazardous waste management plan and management ledger. Step S005: Output the review results and rectification suggestions.

[0029] Basic attribute information, hazard warning information, management and responsibility information, and information-based traceability information may include the following: Basic attribute information includes: name of hazardous waste, waste category and code, form of hazardous waste, main components and hazardous components, date of generation, weight and unit of hazardous waste; Hazard warning information must include at least a hazard icon and the words "Hazardous Characteristics"; Management and accountability information also includes the generating or collecting unit, contact person, and contact information; Information traceability includes hazardous waste digital identification codes and QR codes.

[0030] This invention trains a large model by collecting images of hazardous waste labels. Optionally, the large model is trained with full parameter fine-tuning to establish a large model for cloud-based monitoring of hazardous waste labels, thereby realizing cloud-based monitoring of hazardous waste label compliance.

[0031] Optionally, hazardous waste label data parsing: Large-scale OCR technology is used to automatically parse relevant information from the images of hazardous waste labels. OCR (Optical Character Recognition) is the process of recognizing optical characters.

[0032] Optionally, the content of the discharge permit can be parsed: large-scale OCR technology can be used to automatically parse relevant information from the discharge permit. Optionally, pre-parsing and keyword-based key page location methods can be used to determine the location of the relevant information to be parsed in the discharge permit, thereby reducing the parsing time.

[0033] Optionally, hazardous waste management plans and management ledgers can be retrieved directly from the unit's hazardous waste management plans and management ledgers based on one or more of the information such as unit name, generation date, waste code, and waste name obtained from the hazardous waste label image. Alternatively, the unit's hazardous waste management plans and management ledgers can be retrieved directly from the business platform interface, thus eliminating the need for manual uploading of the enterprise's hazardous waste management plans and management ledgers.

[0034] Cloud-based monitoring big data model: Using big data model algorithms, compliance verification is performed on images of hazardous waste labels, and monitoring reports and rectification opinions are automatically generated to achieve compliance monitoring in the cloud.

[0035] In step S004, the large model reviews the compliance of the comprehensive information of the hazardous waste label based on the input preset rule base, and cross-checks the consistency of the key information of the permit, the hazardous waste management plan, the management ledger and the comprehensive information of the hazardous waste label based on the obtained hazardous waste management plan and management ledger. The large model can conduct the two reviews separately or together. For example, when conducting a cross-review, the compliance of a specific piece of information in the comprehensive information of hazardous waste labeling can be reviewed first, and then the consistency of that information can be cross-reviewed. In other words, there is no order to the reviews.

[0036] The large model reviews the compliance of comprehensive information on hazardous waste labels based on a pre-defined rule base, including: Check whether the hazardous waste label in the image includes a contact person and a contact number, and whether the contact number is an 11-digit mobile phone number or an 11 / 12-digit landline phone number. Check if the hazardous waste label in the image includes the date of generation and if the date of generation is in the correct format; Check whether the hazardous waste label in the image includes the waste form, main components, hazardous components, precautions and remarks. The "remarks" item can be left blank. Check whether the image on the hazardous waste label contains the waste name, waste category, and waste code; Check whether the hazard icons contained in the image are compliant; Check if the hazardous waste label in the image includes the waste weight (and unit); Check whether the hazardous waste label in the image contains a hazardous waste digital identification code, and check whether the hazardous waste digital identification code consists of 37 digits or uppercase English letters; Check if the hazardous waste label in the image contains a compliant QR code.

[0037] A cross-review was conducted to ensure consistency between key information on permits, hazardous waste management plans, management ledgers, and comprehensive information on hazardous waste labels, based on the obtained hazardous waste management plans and management records. Check whether the image on the hazardous waste label contains the waste name, waste category, and waste code. After classifying the waste, check whether the waste code matches the combination of waste category and waste code listed in the "Solid Waste Basic Information Table" of the "Solid Waste Discharge Information" section of the discharge permit. Check whether the waste category, waste code, and waste name match the combination of waste category, waste code, and waste name filed in the "Hazardous Waste Generation Information Table" of the hazardous waste management plan for the current year. Check whether the waste category, waste code, and waste name match the combination of waste category, waste code, and waste name recorded in at least one ledger entry in the hazardous waste management ledger for the current day. Check whether the hazard characteristic icons contained in the image are compliant, and verify whether the hazard characteristic icons are consistent with the hazard characteristic icons corresponding to the waste code (waste code on the hazardous waste label) in the National Hazardous Waste List (Note: If the hazard characteristic icons on the hazardous waste label are in the form of no checkboxes or checkmarks, directly verify the consistency between the hazard characteristic icons on the label and those in the National Hazardous Waste List; if the hazard characteristic icons on the hazardous waste label are in the form of checkboxes and checkmarks, verify the consistency between the hazard characteristic icons with checkmarks on the label and those in the National Hazardous Waste List). Check whether the hazardous waste label in the image contains the waste weight (and unit). Based on the unit name and the date of generation contained in the image, retrieve the hazardous waste management ledger corresponding to the waste code and waste name for that unit on that day, and verify whether any ledger entry has the same waste weight as the label record. Check whether the hazardous waste label in the image contains a hazardous waste digital identification code. Check whether the hazardous waste digital identification code consists of 37 digits or uppercase English letters. Specifically, check whether the first 18 digits are consistent with the unified social credit code stated in the discharge permit, whether the 19th to 26th digits are consistent with the waste code recorded on the hazardous waste label, whether the 27th to 34th digits are consistent with the generation date (yyyymmdd format) recorded on the hazardous waste label, and whether the 35th to 37th digits are all numbers.

[0038] The large model reviews the compliance of comprehensive information on hazardous waste labels based on the input preset rule base, and cross-checks the consistency of key information on permits, hazardous waste management plans, management ledgers and comprehensive information on hazardous waste labels based on the obtained hazardous waste management plans and management ledgers. Output the review results and rectification suggestions.

[0039] In step S005, the review results and rectification suggestions are output. Optionally, the review results can be output item by item in sequence according to the compliance verification items, along with rectification suggestions for non-compliant items: 1. Check if there is a hazardous waste label in the image. If not, the item is non-compliant, and the system will prompt "Please upload the correct hazardous waste label image" and stop outputting the review results for subsequent items; otherwise, the item is compliant, and the review result for item 2 will be output.

[0040] 2. Check if the discharge permit is currently valid. If it is not, prompt "Please upload a valid discharge permit" and stop outputting the review results for subsequent items. Determine if the "Unit Name" stated in the discharge permit matches the "Unit Name (including generating / collecting unit)" contained in the hazardous waste label image. If they do not match, prompt "Please upload a discharge permit that matches the unit on the hazardous waste label". Otherwise, this item is compliant, and output the review results for items 3 to 8.

[0041] 3. Check if the hazardous waste label in the image contains the waste name, waste category, and waste code. If not, this item is non-compliant, and the message will indicate: "According to the 'Technical Specification for Setting Hazardous Waste Identification Marks' (HJ 1276—2022) 5.1.2, 'Hazardous waste labels should include the waste name, waste category, waste code, waste form, hazardous characteristics, main components, harmful components, precautions, name of the generating / collecting unit, contact person, contact information, date of generation, waste weight, and remarks.' Enterprises should accurately fill in the waste category, waste code, waste name, and other information as required, based on the actual hazardous waste generated." Check if the waste category and waste code are consistent with a specific waste category and waste code combination listed in the "Solid Waste Basic Information Table" section of the "Solid Waste Discharge Information" section of the discharge permit. If there is no consistent item, this item is non-compliant, and the message will indicate: "According to the 'Regulations on the Administration of Discharge Permits' (State Council...)..." Article 13 of the State Council Order No. 736 stipulates that the discharge permit shall record the following information: "...location and number of pollutant discharge outlets, pollutant discharge methods and discharge destinations, etc..." and Article 15 stipulates that "the number of pollutant discharge outlets or the type, amount, or concentration of pollutants discharged increases." Enterprises should truthfully declare the hazardous waste generated during actual production and operation and promptly apply for changes to the discharge permit. The waste category, waste code, and waste name must match the combination of waste category, waste code, and waste name filed in the "Hazardous Waste Generation Information Table" of the current year's hazardous waste management plan. If there is no matching item, the item is non-compliant, and the message "According to the 'Technical Guidelines for the Formulation of Hazardous Waste Management Plans and Management Ledgers' (HJ...)" will be displayed. According to Article 5.5.1 of GB / T 1259-2022, the information on hazardous waste generation should meet the following requirements: Hazardous waste name, category, code, and hazardous characteristics: determined and filled in according to the National Hazardous Waste List or GB 5085.1~7 and HJ 298. If there is an industry common name or internal name of the unit, fill in the industry common name or internal name of the unit as well… Enterprises should truthfully record the hazardous waste generated during actual production and operation in their management plans.” The waste category, waste code, and waste name should be consistent with the combination of waste category, waste code, and waste name in at least one record in the hazardous waste generation ledger for that day. If there is no consistency, the item is non-compliant, and the message “According to the Technical Guidelines for the Formulation of Hazardous Waste Management Plans and Management Ledgers (HJ…”) will be displayed. Article 1259—2022, section 6.3.1, requires that "at the generation stage of hazardous waste, the following information should be recorded: batch code, generation time, name of hazardous waste, waste category, waste code, generation quantity, unit of measurement, container / packaging code, container / packaging type, quantity of containers / packaging, hazardous waste generating facility code, responsible person of the generating department, and destination. Enterprises should accurately record the hazardous waste generation ledger based on the actual generation of hazardous waste"; otherwise, this item is compliant.

[0042] 4. Check if the hazard characteristic icons in the image are compliant. If not, this item is non-compliant, and the message will state: "According to the requirements of 5.2.4 of the 'Technical Specification for Setting Hazardous Waste Identification Signs' (HJ 1276—2022), 'Based on the hazardous characteristics of the hazardous waste (including corrosivity, toxicity, flammability, and reactivity), the corresponding hazard characteristic warning graphics in the appendix should be selected and printed on the corresponding position on the label, or printed separately and pasted on the corresponding position on the label. For wastes with multiple hazardous characteristics, all corresponding graphics should be set.' Enterprises should print compliant hazard characteristic icons on the hazardous waste labels according to the hazardous characteristics of the hazardous waste." Check if the hazard characteristic icons contained in (or checked) the image are consistent with the hazard characteristics corresponding to the waste code in the National Hazardous Waste List. If they are inconsistent, this item is non-compliant, and the message will state: "According to the requirements of 5.2.4 of the 'Technical Specification for Setting Hazardous Waste Identification Signs' (HJ 1276—2022), 'Based on the hazardous characteristics of the hazardous waste (including corrosivity, toxicity, flammability, and reactivity), the corresponding hazard characteristic warning graphics in the appendix should be selected and printed on the corresponding position on the label, or printed separately and pasted on the corresponding position on the label. For wastes with multiple hazardous characteristics, all corresponding graphics should be set.' Enterprises should print compliant hazard characteristic icons on the hazardous waste labels according to the hazardous characteristics of the hazardous waste." According to Article 5.2.4 of GB / T 1276—2022, "Based on the hazardous characteristics of hazardous waste (including corrosivity, toxicity, flammability, and reactivity), the corresponding hazard warning graphics from the appendix should be selected and printed on the corresponding position on the label, or printed separately and affixed to the corresponding position on the label. For wastes with multiple hazardous characteristics, all corresponding graphics should be provided." Enterprises should print (or check) the corresponding hazard characteristic icons on the hazardous waste label according to the hazardous characteristics of the waste; otherwise, this item is compliant.

[0043] 5. Check if the hazardous waste label in the image includes the waste weight and unit of measurement. If not, this item is non-compliant, and the message will state: "According to the requirements of 5.1.2 of the 'Technical Specification for Setting Hazardous Waste Identification Signs' (HJ 1276—2022), 'Hazardous waste labels should include waste name, waste category, waste code, waste form, hazardous characteristics, main components, harmful components, precautions, name of generating / collecting unit, contact person, contact information, date of generation, waste weight, and remarks.' Enterprises should accurately fill in the waste weight information as required, based on the actual hazardous waste generated." Check if the hazardous waste label in the image matches the waste weight in any hazardous waste generation ledger corresponding to the waste code and waste name for that day. The message will state: "According to the 'Technical Guidelines for the Development of Hazardous Waste Management Plans and Management Ledgers' (HJ 1276—2022),..." Article 1259—2022, section 6.3.1, requires that "at the generation stage of hazardous waste, the following information should be recorded: batch code, generation time, name of hazardous waste, waste category, waste code, generation quantity, unit of measurement, container / packaging code, container / packaging type, quantity of containers / packaging, hazardous waste generating facility code, responsible person of the generating department, and destination. Enterprises should accurately record the hazardous waste generation ledger based on the actual generation of hazardous waste"; otherwise, this item is compliant.

[0044] 6. Check whether the hazardous waste label in the image includes the waste form, main components, hazardous components, precautions, and remarks (the "remarks" section can be left blank). If at least one item is missing, this item is non-compliant, and the message will read: "According to the requirements of 5.1.2 of the 'Technical Specification for Setting Hazardous Waste Identification Marks' (HJ 1276—2022), 'Hazardous waste labels should include the waste name, waste category, waste code, waste form, hazardous characteristics, main components, hazardous components, precautions, name of the generating / collecting unit, contact person, contact information, date of generation, waste weight, and remarks.' Enterprises should accurately fill in the waste form, main components, hazardous components, precautions, etc., as required, based on the actual hazardous waste generated." Otherwise, this item is compliant.

[0045] 7. Check whether the hazardous waste label in the image contains a compliant QR code. If it does not contain a QR code or the QR code is non-compliant, this item is non-compliant, and the prompt will state: "According to the requirements of 7.1.4 of the 'Technical Specification for Setting Hazardous Waste Identification Signs' (HJ 1276—2022), 'QR codes should be set for the signs of hazardous waste storage, utilization, and disposal facilities to enable information management of facility usage.' In conjunction with the requirements of 5.2.11, 'the digital identification code should be encoded according to the requirements of Article 8 of this standard, achieving 'one item, one code.' The encoded data structure of the hazardous waste label QR code should include the content of the digital identification code, and the information service system should preferably include the information set on the label.' The relevant information should be filled in truthfully according to the actual situation before generating the QR code." Otherwise, this item is compliant.

[0046] 8. Check whether the hazardous waste label in the image contains a 37-digit hazardous waste numerical identification code consisting of numbers or uppercase English letters. If it does not contain a numerical identification code, or if the character length or character type of the numerical identification code is non-compliant, then this item is non-compliant, and the message "According to the 'Technical Specification for Setting Hazardous Waste Identification Marks' (HJ)" will be displayed. According to Article 8.1 of GB / T 1276-2022, the digital identification code on hazardous waste labels consists of 37 digits in four segments: the first segment is the code of the hazardous waste generating or collecting unit (18 digits); the second segment is the waste code (8 digits); the third segment is the generation or collection date code (8 digits); and the fourth segment is the waste sequence code (3 digits). Enterprises should accurately compile the hazardous waste digital identification code based on the unified social credit code recorded in the discharge permit, as well as the waste code, generation date, and sequence code of the batch of hazardous waste. The inspection should check whether the first 18 digits of the 37-digit hazardous waste digital identification code are consistent with the unified social credit code stated in the discharge permit; whether the 19th-26th digits are consistent with the waste code recorded on the hazardous waste label; whether the 27th-34th digits are consistent with the generation date (yyyymmdd format) recorded on the hazardous waste label; and whether the 35th-37th digits are all numbers. If at least one of these is inconsistent or non-compliant, the item is non-compliant, and the message "According to the Technical Specification for Setting Hazardous Waste Identification Marks (HJ)" is displayed. Article 8.1 of GB / T 1276-2022 requires that "the digital identification code on the hazardous waste label consists of 37 digits in 4 segments, of which: the first segment is the code of the hazardous waste generating or collecting unit, 18 digits; the second segment is the waste code, 8 digits; the third segment is the generation or collection date code, 8 digits; and the fourth segment is the waste sequence code, 3 digits. Enterprises should accurately compile the hazardous waste digital identification code based on the unified social credit code recorded in the discharge permit, as well as the waste code, generation date, and sequence code of the batch of hazardous waste." Otherwise, this item is compliant.

[0047] The preset rule base includes rule documents such as the "Technical Specification for Setting Hazardous Waste Identification Signs" and "Environmental Protection Graphic Symbols - Solid Waste Storage (Disposal) Sites".

[0048] This invention can automatically analyze images and extract relevant information from hazardous waste labels, and use large-scale model algorithms for compliance verification, accurately identifying non-compliant items with an accuracy rate of no less than 85%, avoiding inaccurate inspection results due to human error or oversight. Inspectors no longer need to visit the sites of hazardous waste generation, storage, self-utilization / disposal, and transfer to inspect hazardous waste facility signs and discharge permits, avoiding travel delays and reducing inspection time by approximately 90%, significantly lowering inspection time costs. Hazardous waste labels and discharge permits can be uploaded by on-site personnel; inspectors only need to monitor remotely via the cloud, significantly reducing labor costs. It can significantly save inspection time and reduce labor costs, while simultaneously enabling the inspection of multiple hazardous waste labels from multiple companies, significantly improving inspection efficiency.

[0049] This invention solves the technical problems of existing manual on-site inspection methods, such as being time-consuming and labor-intensive, unable to provide comprehensive coverage, lacking timeliness, and requiring highly skilled inspectors. This invention can ensure the accuracy of compliance inspection of hazardous waste labels, while improving inspection efficiency and intelligence.

[0050] In one embodiment, the acquired hazardous waste label images are parsed using a large model to extract comprehensive information about the hazardous waste labels. The large model is trained using images of hazardous waste labels and includes: The image is preprocessed to obtain a preprocessed image. The tensor of the preprocessed image is then input into the OCR visual model. The OCR visual model extracts features from the preprocessed image and generates multiple detection boxes with coordinates. Classify the content within each detection box and provide a confidence score; Filter out detection boxes with confidence scores lower than the preset confidence threshold; The duplicate detection boxes are removed using the nonmaximum suppression algorithm; For each deduplicated detection box, a structured data list is generated, consisting of the category, confidence score, border information, and category name.

[0051] When classifying the content within each detection box, for example, into categories such as "unit name", "waste code" and "hazardous characteristic icon", the confidence score represents the reliability of the detection result, with a preset confidence threshold of, for example, 0.5.

[0052] After recognizing the images in this embodiment, a structured data list is formed, which provides a foundation for subsequent large-scale model review.

[0053] In one embodiment, after the structured data list is formed, the existence of a "hazardous characteristics" label is checked in the structured data list. If it does not exist, the process is terminated, and a message is displayed indicating that the image of the hazardous waste label does not meet the requirements.

[0054] Upon detecting an image error, the process is terminated, and a prompt is sent to upload the correct image, saving time and data traffic.

[0055] In one embodiment, the basic attribute information includes: hazardous waste name, waste category and code, hazardous waste form, main components and harmful components, generation date, hazardous waste weight and unit; Hazard warning information must include at least a hazard icon and the words "Hazardous Characteristics"; Management and accountability information also includes the generating or collecting unit, contact person, and contact information; Information traceability includes hazardous waste digital identification codes and QR codes.

[0056] By identifying the above information, the large model can be properly audited.

[0057] In addition, hazardous waste labels also include precautions, remarks, and other information.

[0058] In one embodiment, when parsing the hazard characteristic icon, it is detected whether there is a checkbox. If so, a pairing relationship is established between the hazard characteristic icon and the adjacent checkbox, and the checked state of the checkbox is identified. Only the hazard characteristic icon corresponding to the checkbox in the checked state is identified as the actual hazard characteristic of the hazardous waste label.

[0059] Reference Figure 2 and Figure 3 Hazardous property icons are basically divided into two types: one without checkboxes and one with checkboxes. For those without checkboxes, it is only necessary to identify the hazardous property icon. For those with checkboxes, it is also necessary to identify whether the checkboxes are checked in order to determine the correct type of hazardous waste represented by the current icon.

[0060] In one embodiment, the key information of the obtained discharge permit is parsed using a large model, including: Combining a pre-set knowledge base, the large model pre-parses the discharge permit, locates the key page numbers in the discharge permit based on keywords in the pre-set knowledge base, converts the key page numbers into images and passes them to the multimodal model for parsing, and uses a pre-set rule base to calibrate the parsing results of the multimodal model. The parsing results include the permit validity period, unit name, unified social credit code, waste category and waste code.

[0061] Optionally, the key page numbers in the discharge permit are located based on keywords in a preset knowledge base, and the key page numbers are converted into images and passed to the multimodal model for parsing, including: Scan the discharge permits to find key page numbers that are semantically related to the keywords in the knowledge base; Convert key page numbers into image format; The key page numbers, converted to image format, are input into the multimodal model, which then extracts parsing information from the key page numbers based on the keywords.

[0062] Optionally, semantic relevance analysis is used to locate page numbers containing key information (such as company name, expiration date, and hazardous waste label); the located key page numbers are converted from PDF documents to high-resolution image format, and image quality enhancement processing is performed to optimize subsequent recognition results; The converted image is input into a multimodal large model, and structured information is extracted by combining targeted parsing prompts, including but not limited to: license validity period (start date and end date), unit name, hazardous waste digital identification code, and other relevant information; The purpose of pre-parsing is not to extract specific field values, but to prepare for detailed parsing: to find the pages and approximate locations containing important information, thereby avoiding detailed parsing of the entire PDF, reducing the processing burden of large models, focusing on key areas, and avoiding noise interference.

[0063] The pre-parsing identifies the locations where the following fields may exist: the basic information area of ​​the discharge permit (the possible locations of the company name, the permit number, and the validity period), the solid waste information area (the table header row, the table data area, and the titles of the hazardous waste-related sections), and format features (the table boundaries and row and column structure, the common locations of field labels (such as labels usually preceding values), and the common ways to write date formats).

[0064] The results found by pre-analysis are as follows: Page 3: May contain "Company Name", confidence level 85%; Page 28: May contain "Hazardous Waste Label Form", confidence level 90%.

[0065] The pre-parsed operations are as follows: Perform high-precision OCR on the specified area (120, 180, 200, 30) on page 3, and perform table structure analysis on the table area on page 28 to extract specific company name strings and facility lists.

[0066] The benefits of pre-parsing: For a 100-page PDF, only 3 pages may contain the necessary information. Scanning all of them would be time-consuming and laborious. Large-scale API calls are billed by token. Pre-parsing reduces unnecessary token usage, allowing tokens to focus on key areas and reducing noise interference.

[0067] Optionally, after parsing the key information, the process also includes determining whether the discharge permit is currently valid. If the discharge permit has expired, the determination process terminates. It further involves determining whether the "unit name" stated in the discharge permit matches the "unit name (including generating or collecting unit)" shown on the hazardous waste label image. If they do not match, the determination process terminates. If the unit name on the discharge permit matches the one on the hazardous waste label, then based on the "Solid Waste Discharge Information" section located on the discharge permit, the "code" in the "Solid Waste Category" column of the "Solid Waste Basic Information Table" is parsed.

[0068] The default knowledge base is a pre-built knowledge base.

[0069] This embodiment adopts a two-stage parsing architecture, an innovative combination of coarse positioning and fine extraction. Instead of a fixed process, it dynamically generates a parsing plan, and the errors in the pre-parsing can be corrected by detailed parsing.

[0070] In one embodiment, a veto rule is set up, and when the veto condition is met, a preliminary verification report is generated for the hazardous waste label; If an image of a hazardous waste label is detected as a questionable label image, it is sent to an expert assistance platform. The expert assistance platform analyzes the questionable label image, marks the violations, uploads the evaluation evidence, and generates a final review report. The questionable label image with the marked violations, the evaluation evidence, and the final review report are used to train the large model.

[0071] A veto rule can be set up, for example, to automatically block obviously non-compliant labels and generate a "Preliminary Verification Report" when there are discrepancies such as inconsistent unit names or missing hazardous waste digital identification codes. For questionable cases (such as unusual layout of hazardous characteristic icons), structured questions are generated using NLP (Natural Language Processing) technology and pushed to the expert collaboration platform, where environmental regulatory experts are responsible for handling them.

[0072] Expert Collaboration Platform: Environmental regulatory experts can view suspicious labeled images, analysis results, and related data (such as management plans and ledgers) via web or mobile devices, mark violations online, and upload supporting evidence (such as screenshots of clauses in HJ 1276—2022). The platform automatically summarizes the opinions of environmental regulatory experts, generates a "Final Review Report," and feeds typical cases back into the large model for training, improving intelligent recognition capabilities.

[0073] Optionally, the expert collaboration platform adopts dynamic task allocation: it uses a "polling + intelligent scheduling" mechanism to allocate review tasks, giving priority to environmental supervision experts who are familiar with the management of solid and hazardous waste in enterprises, thereby shortening the average review time.

[0074] This embodiment can solve the problem of intelligent systems recognizing complex scenarios such as special typesetting labels and ambiguous industry terms, and improve the accuracy of handling difficult cases; typical cases annotated by experts can be used as training data to continuously optimize the performance of large models and form a closed loop of two-way improvement between "human and machine"; the results of manual review can be used as the basis for administrative reconsideration, reducing the probability of enterprise appeals and reducing administrative dispute cases.

[0075] Reference Figure 4 The present invention also proposes an intelligent agent for monitoring the compliance of hazardous waste labeling, employing the method for monitoring the compliance of hazardous waste labeling as described in any of the preceding claims.

[0076] This invention solves the technical problems of existing manual on-site inspection methods, such as being time-consuming and labor-intensive, unable to provide comprehensive coverage, lacking timeliness, and requiring highly skilled inspectors. This invention can ensure the accuracy of compliance inspection of hazardous waste labels, while improving inspection efficiency and intelligence.

[0077] This invention also proposes a system for monitoring compliance with hazardous waste labeling, comprising: Camera module, weighing module, electronic data acquisition module, edge computing module, and cloud monitoring platform; The camera module, weighing module, and electronic data acquisition module are installed at the hazardous waste generation, storage, transfer, utilization, or disposal site. The weighing module weighs the hazardous waste and transmits the weight information to the electronic data acquisition module. The camera module acquires the image of the hazardous waste label and collects the time information, and then sends the image and time information to the data acquisition terminal. The data acquisition terminal transmits data to the edge computing module. The edge computing module uses a large model to parse the images of the acquired hazardous waste labels to obtain comprehensive information about the hazardous waste labels. The large model is trained using images of hazardous waste labels. The comprehensive information about hazardous waste labels includes at least: basic attribute information, hazardous characteristic warning information, management and responsibility information, and information traceability information. The management and responsibility information includes at least the name of the unit that generates or collects the hazardous waste and the date the hazardous waste is generated. Key information of the obtained discharge permits is extracted using a large model; The cloud-based monitoring platform obtains hazardous waste management plans and management ledgers that correspond to the unit's name and the date of hazardous waste generation. The large model reviews the compliance of comprehensive information on hazardous waste labels based on the input preset rule base, and cross-checks the consistency of key information on permits, hazardous waste management plans, management ledgers and comprehensive information on hazardous waste labels based on the obtained hazardous waste management plans and management ledgers. Output the review results and rectification suggestions.

[0078] This invention automatically collects images and associated data from hazardous waste labels using IoT devices (camera module, weighing module, and electronic data acquisition module), performs localized preprocessing using an edge computing module, and then completes in-depth compliance verification through a large model on a cloud-based monitoring platform, thereby improving data collection efficiency and system response speed.

[0079] Camera modules, weighing modules, and electronic data acquisition modules are deployed at the stages of hazardous waste generation, storage, and transfer to automatically capture label images and simultaneously collect real-time data such as generation time and weight.

[0080] Edge computing module preprocessing: The edge computing module performs preliminary analysis on the images of hazardous waste labels, identifies key information (such as company name, waste code, and generation date), and quickly compares it with a locally stored copy of the discharge permit. Obvious non-compliance items (such as the absence of prominent "hazardous characteristics" text, inconsistent company names, etc.) are intercepted in real time, and only questionable data is uploaded to the cloud-based monitoring platform.

[0081] Deep Cloud Verification: The cloud-based monitoring platform's large model receives questionable data uploaded by edge modules and performs multi-dimensional compliance checks by combining hazardous waste management plans, management ledgers, and national lists, generating a final report. If anomalies are found (such as discrepancies between waste codes and lists), rectification instructions are pushed to the enterprise via IoT terminals, and the data on the cloud-based monitoring platform is updated synchronously.

[0082] This invention filters invalid data through an edge computing module, reducing cloud load, improving system response speed, and reducing data transmission pressure; it achieves "second-level" interception of obvious violations, preventing problematic batches from entering the transfer process, and significantly enhancing the timeliness of risk prevention and control; it is compatible with existing IoT smart terminals (such as smart weighbridges and label printers), so enterprises do not need to make large-scale hardware modifications.

[0083] This invention also proposes a system for monitoring compliance with hazardous waste labeling, comprising: A gateway and multiple microservice modules that communicate through the gateway; The microservice module should include at least: a tag parsing module, a license verification module, and a ledger retrieval module; The label parsing module receives images of hazardous waste labels sent from external sources and parses out the complete information of the hazardous waste labels, which is then sent to the ledger retrieval module. The complete information of the hazardous waste labels includes at least: basic attribute information, hazardous characteristic warning information, management and responsibility information, and information traceability information. The management and responsibility information includes at least the name of the unit that generates or collects the hazardous waste and the date the hazardous waste is generated. After the permit verification module obtains the discharge permit and parses out the key information of the permit, it sends it to the ledger retrieval module. The large model reviews the compliance of comprehensive information on hazardous waste labels based on the input preset rule base, and cross-checks the consistency of key information on permits, hazardous waste management plans, management ledgers and comprehensive information on hazardous waste labels based on the obtained hazardous waste management plans and management ledgers. Output the review results and rectification suggestions.

[0084] This invention adopts a microservice architecture to decompose the system into independent modules (tag parsing module, license verification module, and ledger retrieval module), and achieves elastic scaling through containerized deployment, supporting tens of millions of concurrent tag verifications.

[0085] Modular design: The tag parsing module, license verification module, and ledger retrieval module communicate with each other through a gateway (such as an API gateway). Optionally, each module is deployed in a Kubernetes container (also known as Kubernetes) containerized environment, which can dynamically expand the number of instances according to business volume (such as automatically adding parsing nodes during peak monitoring periods).

[0086] Cross-platform compatibility: Supports integration with different enterprise ERP (Enterprise Resource Planning) and MES (Manufacturing Execution System) systems, obtaining management plans and ledger data through standardized interfaces (such as the HJ 212-2017 protocol). Provides a mobile SDK (Software Development Kit), allowing enterprises to directly upload label images and receive rectification feedback via a mobile app, achieving a "one-stop" operation.

[0087] Intelligent load balancing: It uses distributed message queues (such as Kafka - a distributed stream processing platform) to buffer tasks to be processed, and combines machine learning to predict the load of each module, dynamically allocate computing resources, and improve system throughput.

[0088] This invention enables a single cluster to support concurrent access from 100,000+ enterprises, meeting the large-scale application needs of provincial environmental protection platforms; the failure of any module does not affect the operation of other modules, and the automatic restart mechanism achieves 99.99% service availability; enterprises can choose to subscribe to service modules according to their own needs (such as using only the tag parsing function), reducing the cost of digital transformation.

[0089] Hazardous waste: Solid waste that is listed in the National Hazardous Waste List or identified as having hazardous characteristics according to the national hazardous waste identification standards and methods.

[0090] Hazardous waste labels: These are labels placed on containers or packaging containing hazardous waste. They consist of a combination of text, codes, and graphic symbols and are used to convey specific information about hazardous waste to relevant individuals, serving as a warning of the potential environmental hazards of hazardous waste.

[0091] The above description is merely the principle and preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several other modifications can be made based on the principle of the present invention, and these modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for monitoring compliance with hazardous waste labeling, characterized in that, include: The images of the acquired hazardous waste labels are parsed using a large model to extract comprehensive information about the hazardous waste labels. The large model is trained using images of hazardous waste labels. The comprehensive information about the hazardous waste labels includes at least: basic attribute information, hazardous characteristic warning information, management and responsibility information, and information traceability information. The management and responsibility information includes at least the name of the unit that generates or collects the hazardous waste and the date the hazardous waste is generated. Key information of the obtained discharge permits is extracted using a large model; Obtain the hazardous waste management plan and management ledger that correspond to the organization's name and the date the hazardous waste was generated; The large model reviews the compliance of comprehensive information on hazardous waste labels based on the input preset rule base, and cross-checks the consistency of key information on permits, hazardous waste management plans, management ledgers and comprehensive information on hazardous waste labels based on the obtained hazardous waste management plans and management ledgers. Output the review results and rectification suggestions.

2. The method for monitoring the compliance of hazardous waste labeling according to claim 1, characterized in that, The images of acquired hazardous waste labels are parsed using a large model to extract comprehensive information about the hazardous waste labels. This large model is trained using images of hazardous waste labels and includes: The image is preprocessed to obtain a preprocessed image. The tensor of the preprocessed image is then input into the OCR visual model. The OCR visual model extracts features from the preprocessed image and generates multiple detection boxes with coordinates. Classify the content within each detection box and provide a confidence score; Filter out detection boxes with confidence scores lower than the preset confidence threshold; The duplicate detection boxes are removed using the nonmaximum suppression algorithm; For each deduplicated detection box, a structured data list is generated, consisting of the category, confidence score, border information, and category name.

3. The method for monitoring the compliance of hazardous waste labeling according to claim 2, characterized in that, After the structured data list is generated, check whether the "hazardous characteristics" label exists in the structured data list. If it does not exist, the process is terminated and a message is displayed indicating that the image of the hazardous waste label does not meet the requirements.

4. The method for monitoring the compliance of hazardous waste labeling according to claim 1, characterized in that, Basic attribute information includes: name of hazardous waste, waste category and code, form of hazardous waste, main components and hazardous components, date of generation, weight and unit of hazardous waste; Hazard warning information must include at least a hazard icon and the words "hazardous characteristic"; Management and accountability information also includes the generating or collecting unit, contact person, and contact information; Information traceability includes hazardous waste digital identification codes and QR codes.

5. The method for monitoring the compliance of hazardous waste labeling according to claim 4, characterized in that, When parsing hazard characteristic icons, check if there are checkboxes. If so, establish a pairing relationship between the hazard characteristic icon and the adjacent checkbox, and identify the checked state of the checkbox. Only the hazard characteristic icon corresponding to the checkbox in the checked state is identified as the actual hazard characteristic of the hazardous waste label.

6. The method for monitoring compliance of hazardous waste labeling according to claim 1, characterized in that, The key information of the obtained discharge permits was extracted using a large model, including: Combining a pre-set knowledge base, the large model pre-parses the discharge permit, locates the key page numbers in the discharge permit based on keywords in the pre-set knowledge base, converts the key page numbers into images and passes them to the multimodal model for parsing, and uses a pre-set rule base to calibrate the parsing results of the multimodal model. The parsing results include the permit validity period, unit name, unified social credit code, waste category and waste code.

7. The method for monitoring the compliance of hazardous waste labeling according to claim 1, characterized in that, A veto rule is set up so that a preliminary verification report is generated for the hazardous waste label when the veto conditions are met. If an image of a hazardous waste label is detected as a questionable label image, it is sent to an expert assistance platform. The expert assistance platform analyzes the questionable label image, marks the violations, uploads the evaluation evidence, and generates a final review report. The questionable label image with the marked violations, the evaluation evidence, and the final review report are used to train the large model.

8. An intelligent agent for monitoring the compliance of hazardous waste labeling, characterized in that, The method for monitoring compliance with hazardous waste labeling as described in any one of claims 1-7 is adopted.

9. A system for monitoring compliance with hazardous waste labeling, characterized in that, include: Camera module, weighing module, electronic data acquisition module, edge computing module, and cloud monitoring platform; The camera module, weighing module, and electronic data acquisition module are installed at the hazardous waste generation, storage, transfer, utilization, or disposal site. The weighing module weighs the hazardous waste and transmits the weight information to the electronic data acquisition module. The camera module acquires the image of the hazardous waste label and collects the time information, and then sends the image and time information to the data acquisition terminal. The data acquisition terminal transmits data to the edge computing module. The edge computing module analyzes the acquired hazardous waste label images using a large model to extract comprehensive information about the hazardous waste labels. The large model is trained using images of hazardous waste labels. The comprehensive information about hazardous waste labels includes at least: basic attribute information, hazardous characteristic warning information, management and responsibility information, and information traceability information. The management and responsibility information includes at least the name of the generating or collecting unit. Key information of the obtained discharge permits is extracted using a large model; The cloud-based monitoring platform obtains hazardous waste management plans and management ledgers that correspond to the unit's name and the date of hazardous waste generation. The large model reviews the compliance of comprehensive information on hazardous waste labels based on the input preset rule base, and cross-checks the consistency of key information on permits, hazardous waste management plans, management ledgers and comprehensive information on hazardous waste labels based on the obtained hazardous waste management plans and management ledgers. Output the review results and rectification suggestions.

10. A system for monitoring compliance with hazardous waste labeling, characterized in that, include: A gateway and multiple microservice modules that communicate through the gateway; The microservice module should include at least: a tag parsing module, a license verification module, and a ledger retrieval module; The label parsing module receives images of hazardous waste labels sent from external sources and parses out the complete information of the hazardous waste labels, which is then sent to the ledger retrieval module. The complete information of the hazardous waste labels includes at least: basic attribute information, hazardous characteristic warning information, management and responsibility information, and information traceability information. The management and responsibility information includes at least the name of the unit that generates or collects the hazardous waste and the date the hazardous waste is generated. After the permit verification module obtains the discharge permit and parses out the key information of the permit, it sends it to the ledger retrieval module. The large model reviews the compliance of comprehensive information on hazardous waste labels based on the input preset rule base, and cross-checks the consistency of key information on permits, hazardous waste management plans, management ledgers and comprehensive information on hazardous waste labels based on the obtained hazardous waste management plans and management ledgers. Output the review results and rectification suggestions.

Images