Smart-ws garbage classification intelligent platform
By using multimodal fusion recognition and regional adaptation technologies, the recognition accuracy of the intelligent waste sorting platform has been improved in complex scenarios and regional adaptability. This solves the recognition problem of existing platforms under complex environments and regional differences, and achieves efficient waste sorting management.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANDONG HAIWO JIAMEI ENVIRONMENTAL ENG CO LTD
- Filing Date
- 2026-04-08
- Publication Date
- 2026-07-03
AI Technical Summary
The existing SMART-WS waste sorting digital platform has low recognition accuracy and high misjudgment rate in complex scenarios, and cannot adapt to the differences in waste types in different regions, resulting in insufficient efficiency and standardization of waste sorting management.
The system employs a multimodal fusion recognition module, a regionalized waste feature database, a location matching module, an edge computing module, and a cloud platform. Combined with an improved YOLOv8 model and an anti-interference processing module, it achieves dynamic optimization and regional adaptation of waste features through a multimodal feature self-learning optimization unit.
In complex scenarios, the accuracy of garbage identification has been improved to 97%, and the false judgment rate has been reduced by 70%. The platform has achieved autonomous learning and iterative optimization, adapting to changes in garbage types in different areas without the need for manual intervention, thus reducing operation and maintenance costs.
Smart Images
Figure CN122335281A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of digital technology for waste sorting, specifically the SMART-WS digital platform for waste sorting. Background Technology
[0002] With the comprehensive advancement of waste sorting in my country, digital technologies have become a core support for improving waste sorting efficiency, standardizing waste sorting management, and reducing manual operation and maintenance costs. Currently, most existing SMART-WS-type waste sorting digital platforms and similar products rely on a single AI visual recognition technology to identify and distinguish waste categories. In an ideal laboratory environment, the recognition accuracy can reach over 90%, but in actual outdoor or complex usage scenarios, the recognition effect drops significantly due to various interference factors, making it difficult to meet practical application needs. Existing technology has technical defects: Poor adaptability to complex scenarios and high misjudgment rate. In harsh environments such as rain, night, strong light, and backlight, the clarity of garbage images acquired by image acquisition equipment drops significantly, resulting in problems such as blurriness, glare, and excessive shadows. At the same time, when garbage is covered with oil, obscured by foreign objects such as plastic bags and branches, or when various types of garbage are stuck together or mixed together, the AI visual recognition unit cannot accurately extract the core features of the garbage, leading to recognition errors and missed judgments. According to actual tests, the recognition accuracy of existing platforms drops sharply in such scenarios, seriously affecting the standardization and efficiency of garbage classification management. The existing platform suffers from insufficient regional adaptability, failing to match the diverse waste types across different regions. Significant differences exist in residents' lifestyles, industrial structures, and environmental conditions across regions, leading to variations in waste types and compositions. For instance, urban communities primarily consist of kitchen waste, recyclable daily necessities (plastics, paper, glass, etc.), and a small amount of hazardous waste, while rural villages contain substantial agricultural waste (straw, rice husks, poultry manure, etc.). The current platform lacks targeted adaptation to the specific characteristics of waste from different regions, employing a uniform identification standard and dataset. This results in a high misclassification rate for representative waste from various regions, failing to meet the diverse needs of waste sorting.
[0003] Therefore, we proposed the SMART-WS intelligent waste sorting platform. Summary of the Invention
[0004] The purpose of this invention is to provide a SMART-WS intelligent waste sorting platform, which solves the problems mentioned in the background art.
[0005] To achieve the above objectives, the present invention provides the following technical solution: the SMART-WS waste sorting digital intelligence platform, comprising a multimodal fusion recognition module, a regionalized waste feature database, a location matching module, an edge computing module, a cloud platform, and a feedback optimization module; the multimodal fusion recognition module, the regionalized waste feature database, the location matching module, the edge computing module, the cloud platform, and the feedback optimization module work collaboratively to form a fully closed-loop digital intelligence management system of collection-processing-matching-recognition-feedback-learning-optimization; The multimodal fusion recognition module includes an AI visual recognition unit, an anti-interference processing module, an odor recognition unit, a spectral recognition unit, a weight recognition unit, and a multimodal fusion optimization module; the multimodal fusion optimization module includes an adaptive dynamic weighted fusion unit, a multidimensional feature correlation analysis unit, an environmental adaptive correction unit, and a multimodal feature self-learning optimization unit; The multimodal feature self-learning optimization unit is used to collect feature data, recognition results, environmental parameters and user error correction feedback data of each recognition unit in real time. It learns the feature change patterns of different scenarios and different types of waste through deep learning algorithms, dynamically optimizes feature extraction algorithms, weighting coefficient allocation rules, feature association rules and environmental correction parameters, and automatically identifies new types of waste and extracts their multi-dimensional features to supplement the regional waste feature database.
[0006] In a preferred embodiment of the present invention, the AI visual recognition unit adopts an improved YOLOv8 deep learning model, and the model training set includes a garbage sample dataset of rainy days, nighttime, strong light, oil stains, occlusion, and garbage sticking / mixing scenarios; the AI visual recognition unit enhances the extraction of garbage main features through an attention mechanism, suppresses background interference, and improves the visual recognition accuracy in complex scenarios.
[0007] In a preferred embodiment of the present invention, the anti-interference processing module includes an image enhancement unit, an occlusion segmentation unit, and an adhesion separation unit. The image enhancement unit employs dehazing, adaptive lighting, and oil removal algorithms to specifically process the collected garbage images in order to improve image clarity. The occlusion segmentation unit uses the U-Net image segmentation algorithm to identify and segment occlusion objects and garbage subjects in the image, and extract the garbage subject region. The adhesion separation unit uses contour extraction combined with edge detection algorithm to segment the adhered / mixed waste and extract the feature information of individual waste.
[0008] In a preferred embodiment of the present invention, the adaptive dynamic weighted fusion unit constructs a single waste feature evaluation model, and scores the clarity of four types of identification features—visual, odor, spectral, and weight—on a scale of 0-1. The higher the score, the higher the weighting coefficient. Furthermore, a preset feature priority rule is established: when the score of a certain dimension feature is ≥0.8, its weighting coefficient is automatically increased to 0.5-0.6, while the weights of other dimensions are proportionally reduced to highlight the advantageous identification dimensions.
[0009] In a preferred embodiment of the present invention, the multi-dimensional feature association analysis unit constructs a multi-dimensional feature association database, mines and stores the inherent association rules of visual-odor-spectrum-weight for different waste categories, and forms a feature association map. When there are feature conflicts in the output results of each identification unit, the feature association map is called to verify the association of features in each dimension, and the identification results that conform to the association rules are adopted first, or a comprehensive judgment is made by sorting the credibility of spectrum > odor > weight > visual.
[0010] In a preferred embodiment of the present invention, the environmental adaptive correction unit includes an environmental parameter monitoring submodule and a dynamic correction submodule; The environmental parameter monitoring submodule collects data on ambient humidity, temperature, light intensity, and surrounding odor concentration. In the dynamic correction submodule, the odor recognition unit first removes odor interference through an environmental odor filtering algorithm and then extracts feature parameters; when the spectral recognition unit detects a light intensity ≥10000 lux or a temperature ≥35℃, it automatically starts the dynamic calibration algorithm of the spectral curve to correct the spectral curve offset.
[0011] As a preferred embodiment of the present invention, the regionalized waste feature database adopts a combination of administrative region division and scene type division to store multi-dimensional features and feature association rules of waste in different regions and different scenes. The database is dynamically updated and expanded through on-site data collection, user feedback, sanitation data statistics, data sharing from waste disposal companies, and autonomous identification and supplementation by a multimodal feature self-learning optimization unit.
[0012] In a preferred embodiment of the present invention, the positioning and matching module adopts Beidou and 5G dual-mode positioning technology; through hierarchical matching logic of administrative region, scene type and specific disposal point, it accurately retrieves the garbage feature data of the corresponding area and the rules optimized by the multimodal feature self-learning optimization unit; when the positioning signal is interrupted, the edge computing module calls the most recently retrieved area data for temporary identification, and automatically updates it synchronously after the signal is restored.
[0013] In a preferred embodiment of the present invention, the edge computing module is configured with an edge computing gateway to realize local data preprocessing, real-time identification and device operation status monitoring. The local data preprocessing includes data cleaning, feature extraction, and interference filtering; the device operation status monitoring is used to detect the working status of cameras, sensors, and positioning modules, and sends fault warning signals to the cloud platform and operation and maintenance terminal when a fault occurs; when the network is interrupted, the edge computing module can independently complete the identification, and automatically synchronize local data after the network is restored.
[0014] As a preferred embodiment of the present invention, the cloud platform adopts a hybrid storage architecture of MySQL, MongoDB and Hadoop to realize encrypted data storage, multi-dimensional statistical analysis, dynamic algorithm optimization and multi-terminal collaboration. The feedback optimization module collects user error correction feedback and operation and maintenance inspection data, providing training data for the multimodal feature self-learning optimization unit, forming a closed loop of recognition-feedback-learning-optimization, and promoting the platform's continuous autonomous iteration; The multi-terminal collaboration includes data synchronization and sharing among the regulatory, operational, resident, and waste collection ends, enabling intelligent digital management throughout the entire process.
[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention, through the collaboration of an improved YOLOv8 model and an anti-interference module, combined with scene feature optimization by a self-learning unit, completely breaks through the bottleneck of recognition in complex scenes such as rainy days, nights, and strong light, and is suitable for various outdoor and indoor scenes; combined with a regional database and a positioning module, it accurately matches the characteristics of garbage in different regions. The multimodal feature self-learning optimization unit and the multimodal fusion optimization module (dynamic weighting, feature association, and environment correction) work together in a deep synergy, making the optimization of dynamic weighting, feature association, and environment correction more targeted. The accuracy of complex scene recognition is improved to over 97%, and the false judgment rate is reduced by 70%. Combined with regional databases, the database can be updated autonomously and dynamically, further improving the real-time performance and accuracy of regional adaptation, and solving the pain point of lagging regional adaptation in existing platforms. The self-learning unit works in conjunction with edge computing and feedback optimization modules to enable the platform to learn, iterate, and adapt autonomously. It can adapt to changes in waste types and scenarios without human intervention, significantly reducing the cost of manual operation and maintenance and database updates. At the same time, it can autonomously identify new waste and expand the platform's adaptability. Attached Figure Description
[0016] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart of the SMART-WS intelligent waste sorting platform of the present invention. Detailed Implementation
[0017] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.
[0018] The technical architecture, functional modules, and practical applications of the SMART-WS waste sorting digital platform are described in detail. This platform is an integrated solution for the digital management of waste sorting. At its core, it solves the industry pain points of traditional waste sorting identification platforms, such as poor adaptability to complex scenarios and insufficient regional adaptability, through multimodal fusion recognition and regional precision adaptation technology.
[0019] The platform comprises six core modules: a multimodal fusion recognition module, a regionalized waste feature database, a location matching module, an edge computing module, a cloud platform, and a feedback optimization module. It constructs a fully closed-loop digital intelligent management system encompassing collection, processing, matching, recognition, feedback, learning, and optimization, achieving high accuracy in waste identification, high regional adaptability, and autonomous iterative optimization. In complex scenarios, the waste identification accuracy can reach over 97%, with a 70% reduction in false judgment rate. Simultaneously, it supports multi-terminal collaborative management from regulatory, operational, resident, and collection ends, covering the entire process of waste sorting, identification, supervision, and collection.
[0020] This platform can be widely used in waste sorting and disposal points in various scenarios such as urban communities, rural villages, industrial parks, shopping malls, and scenic spots. Through the integrated application of technologies such as Beidou and 5G dual-mode positioning, edge computing, and deep learning, it achieves a combination of local real-time identification and cloud-based intelligent management. When the positioning signal or network is interrupted, it can independently complete the core identification work and automatically synchronize the data after recovery. It meets the actual application needs of various complex outdoor and indoor environments, significantly reduces the manual operation and maintenance costs of waste sorting, and improves the standardization and intelligence level of waste sorting management.
[0021] Platform hardware deployment and initial configuration Hardware equipment is deployed at each waste sorting and disposal point, including visual acquisition cameras, odor sensors, spectral sensors, weight sensors, Beidou and 5G dual-mode positioning terminals, edge computing gateways, and environmental parameter monitoring sensors (temperature, humidity, light intensity, odor concentration). Communication connections between each device and the edge computing gateway are established to ensure that the hardware devices can collect data and transmit instructions normally.
[0022] The initial setup of the cloud platform was completed, and a cloud data storage and analysis system was built based on a hybrid storage architecture of MySQL, MongoDB, and Hadoop. Basic functions such as data encryption, multi-terminal permission management, and algorithm deployment were configured. At the same time, the initial data entry of the regional waste feature database was completed. By collecting multi-dimensional features (visual, odor, spectrum, weight) of waste from various administrative regions and scenario types (communities, rural areas, industrial parks, etc.) on-site, and combining the shared data of sanitation departments and waste disposal companies, an initial feature library and feature association rules were established.
[0023] The algorithm model of the multimodal fusion recognition module is initially deployed. The improved YOLOv8 visual recognition model, U-Net image segmentation algorithm, dynamic weighted fusion algorithm, and spectral curve calibration algorithm are deployed to the edge computing gateway and cloud platform. The parameters of each algorithm model are initialized, and core rules such as feature priority rules, feature credibility ranking rules (spectral > odor > weight > vision) and environmental correction trigger thresholds (light ≥ 10000 lux, temperature ≥ 35℃) are preset.
[0024] Complete the hierarchical matching logic configuration of the location matching module, and set three-level retrieval rules for administrative region, scenario type, and specific disposal point to ensure that the location terminal of each disposal point can accurately associate with the corresponding data in the regional waste feature database.
[0025] Complete the initialization of the multi-terminal collaborative system, configure corresponding accounts and operation permissions for the regulatory end, operation end, resident end, and waste collection end, and develop data synchronization interfaces between each terminal and the cloud platform.
[0026] Platform core operation implementation steps and process The core workflow of this platform revolves around data collection, preprocessing, localization and matching, multimodal fusion recognition, cloud synchronization, feedback optimization, and self-learning. Each module works collaboratively to form a closed-loop operation. The specific steps are as follows: 1. Data Acquisition Stage: Various sensors and acquisition devices at the disposal point collect waste-related data and environmental parameters in real time. Among them, the AI visual recognition unit's camera collects waste image data, the odor sensor collects waste odor characteristic data, the spectral sensor collects waste spectral characteristic data, and the weight sensor collects waste weight data. The environmental parameter monitoring sensor simultaneously collects environmental data such as humidity, temperature, light intensity, and surrounding odor concentration at the disposal point. All collected data is transmitted to the edge computing gateway in real time.
[0027] 2. Local Data Preprocessing Stage: The edge computing module performs local preprocessing on the collected raw data, mainly completing three tasks: ① The anti-interference processing module optimizes the garbage images. The image enhancement unit improves image clarity through defogging, adaptive lighting, and oil removal algorithms. The occlusion segmentation unit segments occluders from the main garbage body using the U-Net algorithm. The adhesion separation unit segments adhered / mixed garbage through contour extraction and edge detection algorithms, extracting visual features of individual garbage. ② Data cleaning and interference filtering are performed on odor, spectral, and weight data to remove abnormal and invalid data. ③ Effective feature parameters of garbage in each dimension are extracted to prepare for subsequent identification.
[0028] 3. Location Matching Stage: The location matching module obtains the precise location information of the current disposal point through Beidou and 5G dual-mode positioning technology. According to the hierarchical matching logic of administrative region, scenario type and specific disposal point, it accurately retrieves the corresponding regional and scenario waste feature data and feature association rules from the regionalized waste feature database of the cloud platform. If the positioning signal is interrupted, the edge computing module directly calls the most recently retrieved regional data for temporary identification. After the signal is restored, it automatically synchronizes the latest data of the cloud platform.
[0029] 4. Multimodal Fusion Recognition Stage: Based on the preprocessed feature data and retrieved regional feature rules, the edge computing module initiates the multimodal fusion recognition module to complete waste category identification, which consists of 4 sub-steps: Step 4.1: Environmental adaptive correction. The environmental adaptive correction unit dynamically corrects the odor and spectral characteristics based on the collected environmental parameters. The odor recognition unit removes odor interference through an environmental odor filtering algorithm. If the light intensity is ≥10000 lux or the temperature is ≥35℃, the spectral recognition unit activates the spectral curve dynamic calibration algorithm to correct the spectral offset. Step 4.2: Feature clarity scoring. The adaptive dynamic weighted fusion unit uses a single waste feature evaluation model to score the clarity of four types of features—visual, odor, spectral, and weight—on a 0-1 scale. Step 4.3: Dynamic weighting allocation. Assign weighting coefficients to each feature based on the scoring results. The higher the score, the higher the weight. If the score of a certain dimension feature is ≥0.8, increase its weighting coefficient to 0.5-0.6, and adjust the weights of other dimensions proportionally. Step 4.4: Feature association verification and comprehensive judgment. The multi-dimensional feature association analysis unit calls the feature association map to verify the correlation of the output results of each identification unit. If there is no conflict in the results, the waste category is directly output. If there is a feature conflict, the results that conform to the association rules are adopted first, or the credibility is sorted by spectrum > odor > weight > visual for comprehensive judgment. Finally, the real-time identification of the waste category is completed locally, and the identification results are pushed to the display terminal at the disposal point to guide residents to dispose of waste correctly.
[0030] 5. Cloud Synchronization and Multi-Terminal Collaboration Stage: The edge computing module synchronizes local waste identification data, equipment operation status data, and environmental parameter data to the cloud platform in real time. The cloud platform performs encrypted storage and multi-dimensional statistical analysis on the data (such as the amount of waste disposed of at each disposal point, the proportion of different categories, and the accuracy of identification), and synchronizes the core data to the supervision end, operation end, resident end, and collection end: the supervision end can view the waste classification management data of the entire area, the operation end can view the equipment operation status, the resident end can view their own disposal records, and the collection end can realize intelligent scheduling based on the waste disposal data.
[0031] 6. Fault Monitoring and Early Warning Phase: The edge computing module monitors the working status of cameras, various sensors, and positioning modules in real time. If a device fault is detected, a fault warning signal is immediately sent to the cloud platform and the operation and maintenance terminal. Operation and maintenance personnel can perform timely equipment maintenance based on the warning information to ensure the continuous normal operation of the platform. If the network is interrupted, the edge computing module independently completes garbage identification and local data storage. After the network is restored, the local data is automatically synchronized to the cloud platform without data loss.
[0032] 7. Feedback Data Collection Phase: The feedback optimization module collects two types of core feedback data in real time. The first is user error correction feedback. If residents find that the identification result is incorrect, they can submit error correction information (including the actual type of waste, image, etc.) through the terminal at the disposal point or the resident's mobile app. The second is operation and maintenance inspection data. Operation and maintenance personnel submit data such as equipment maintenance and on-site verification of waste identification problems to the platform. All feedback data is uploaded to the cloud platform to provide training data for the platform's self-learning.
[0033] 8. Autonomous Learning and Optimization Phase: The multimodal feature self-learning optimization unit collects feature data, recognition results, environmental parameters, and all feedback data from each recognition unit in real time. Through deep learning algorithms, it autonomously learns the feature variation patterns of different scenarios and types of waste, completing four core optimizations: ① Dynamically optimizing feature extraction algorithms for visual, odor, and other dimensions; ② Optimizing the weighting coefficient allocation rules of the adaptive dynamic weighted fusion unit; ③ Optimizing the feature association rules of the multi-dimensional feature association analysis unit; ④ Optimizing the environmental correction parameters of the environmental adaptive correction unit. Simultaneously, this unit can autonomously identify newly added waste types, automatically extract their multi-dimensional features, and supplement them to the regionalized waste feature database, achieving autonomous dynamic updates to the database.
[0034] 9. Algorithm and Rule Synchronization Phase: The cloud platform will synchronize the self-learning optimized algorithm model, feature rules, and correction parameters to the edge computing gateways of all disposal points, complete the algorithm and rule updates of the entire platform, realize the platform's autonomous iteration and full-domain adaptation, and adapt to the updates of waste types and changes in scenarios without manual intervention, continuously improving the recognition accuracy and regional adaptability.
[0035] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or basic characteristics. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of equivalents of the claims be included within the present invention.
[0036] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.
Claims
1. A smart waste sorting intelligent platform, characterized in that: It includes a multimodal fusion recognition module, a regionalized waste feature database, a location matching module, an edge computing module, a cloud platform, and a feedback optimization module; the multimodal fusion recognition module, the regionalized waste feature database, the location matching module, the edge computing module, the cloud platform, and the feedback optimization module work together to form a closed-loop digital intelligent management system that includes collection, processing, matching, recognition, feedback, learning, and optimization; The multimodal fusion recognition module includes an AI visual recognition unit, an anti-interference processing module, an odor recognition unit, a spectral recognition unit, a weight recognition unit, and a multimodal fusion optimization module; the multimodal fusion optimization module includes an adaptive dynamic weighted fusion unit, a multidimensional feature correlation analysis unit, an environmental adaptive correction unit, and a multimodal feature self-learning optimization unit; The multimodal feature self-learning optimization unit is used to collect feature data, recognition results, environmental parameters and user error correction feedback data of each recognition unit in real time. It learns the feature change patterns of different scenarios and different types of waste through deep learning algorithms, dynamically optimizes feature extraction algorithms, weighting coefficient allocation rules, feature association rules and environmental correction parameters, and automatically identifies new types of waste and extracts their multi-dimensional features to supplement the regional waste feature database.
2. The SMART-WS intelligent waste sorting platform according to claim 1, characterized in that: The AI visual recognition unit adopts an improved YOLOv8 deep learning model. The model training set includes garbage sample datasets in rainy weather, nighttime, strong light, oil stains, occlusion, and garbage sticking / mixing scenarios. The AI visual recognition unit enhances the extraction of main features of garbage through an attention mechanism, suppresses background interference, and improves the accuracy of visual recognition in complex scenarios.
3. The SMART-WS intelligent waste sorting platform according to claim 1, characterized in that: The anti-interference processing module includes an image enhancement unit, an occlusion segmentation unit, and an adhesion separation unit. The image enhancement unit employs dehazing, adaptive lighting, and oil removal algorithms to specifically process the collected garbage images in order to improve image clarity. The occlusion segmentation unit uses the U-Net image segmentation algorithm to identify and segment occlusion objects and garbage subjects in the image, and extract the garbage subject region. The adhesion separation unit uses contour extraction combined with edge detection algorithm to segment the adhered / mixed waste and extract the feature information of individual waste.
4. The SMART-WS intelligent waste sorting platform according to claim 1, characterized in that: The adaptive dynamic weighted fusion unit constructs a single waste feature evaluation model, which scores the clarity of four types of identification features—visual, odor, spectral, and weight—on a scale of 0-1. The higher the score, the higher the weighting coefficient. Furthermore, a preset feature priority rule is established: when the score of a certain dimension feature is ≥0.8, its weighting coefficient is automatically increased to 0.5-0.6, while the weights of other dimensions are proportionally reduced to highlight the advantageous identification dimensions.
5. The SMART-WS intelligent waste sorting platform according to claim 1, characterized in that: The multi-dimensional feature association analysis unit constructs a multi-dimensional feature association database, mines and stores the inherent association rules of visual-odor-spectrum-weight for different waste categories, and forms a feature association map; When there are feature conflicts in the output results of each recognition unit, the feature association map is called to verify the correlation of features in each dimension. The recognition results that conform to the association rules are adopted first, or a comprehensive judgment is made by sorting the credibility by spectrum > odor > weight > vision.
6. The SMART-WS waste sorting digital platform according to claim 1, characterized in that: The environmental adaptive correction unit includes an environmental parameter monitoring submodule and a dynamic correction submodule; The environmental parameter monitoring submodule collects data on ambient humidity, temperature, light intensity, and surrounding odor concentration. In the dynamic correction submodule, the odor recognition unit first removes odor interference through an environmental odor filtering algorithm and then extracts feature parameters; when the spectral recognition unit detects a light intensity ≥10000 lux or a temperature ≥35℃, it automatically starts the dynamic calibration algorithm of the spectral curve to correct the spectral curve offset.
7. The SMART-WS waste sorting digital platform according to claim 1, characterized in that: The regionalized waste feature database adopts a combination of administrative region division and scene type division to store multi-dimensional features and feature association rules of waste in different regions and different scenes. The database is dynamically updated and expanded through on-site data collection, user feedback, sanitation data statistics, data sharing from waste disposal companies, and autonomous identification and supplementation by a multimodal feature self-learning optimization unit.
8. The SMART-WS waste sorting digital platform according to claim 1, characterized in that: The positioning and matching module adopts Beidou and 5G dual-mode positioning technology; through hierarchical matching logic of administrative region, scene type and specific disposal point, it accurately retrieves the garbage feature data of the corresponding area and the rules optimized by the multimodal feature self-learning optimization unit; when the positioning signal is interrupted, the edge computing module calls the most recently retrieved area data for temporary identification, and automatically updates it synchronously after the signal is restored.
9. The SMART-WS intelligent waste sorting platform according to claim 1, characterized in that: The edge computing module is configured with an edge computing gateway to realize local data preprocessing, real-time identification, and device operation status monitoring. The local data preprocessing includes data cleaning, feature extraction, and interference filtering; the device operation status monitoring is used to detect the working status of cameras, sensors, and positioning modules, and sends fault warning signals to the cloud platform and operation and maintenance terminal when a fault occurs; when the network is interrupted, the edge computing module can independently complete the identification, and automatically synchronize local data after the network is restored.
10. The SMART-WS waste sorting digital platform according to claim 1, characterized in that: The cloud platform adopts a hybrid storage architecture of MySQL, MongoDB and Hadoop to achieve encrypted data storage, multi-dimensional statistical analysis, dynamic algorithm optimization and multi-terminal collaboration; The feedback optimization module collects user error correction feedback and operation and maintenance inspection data, providing training data for the multimodal feature self-learning optimization unit, forming a closed loop of recognition-feedback-learning-optimization, and promoting the platform's continuous autonomous iteration; The multi-terminal collaboration includes data synchronization and sharing among the regulatory, operational, resident, and waste collection ends, enabling intelligent digital management throughout the entire process.