Garbage classification recycling intelligent management system based on Internet of Things

Abstract

The invention provides a garbage classification and recovery intelligent management system based on the Internet of Things, and relates to the technical field of garbage classification, and the system comprises a data collection module which is used for collecting physical information, image information and environment information of thrown garbage in real time; the feature extraction module is used for extracting physical features, image features and pollution features; the garbage type identification module is used for identifying the type of thrown garbage and marking abnormal garbage; the garbage classification module is used for classifying the thrown garbage; the abnormity detection module is used for analyzing abnormity reasons; the exception handling module is used for implementing different handling schemes on the abnormal garbage; and the abnormity tracing module is used for tracing the abnormal garbage. According to the method, deep fusion of multi-source heterogeneous data is realized by adopting multi-model fusion identification; the classification result is dynamically verified by introducing a confidence evaluation mechanism, so that the false drop rate is effectively reduced, and the accuracy and robustness of garbage classification in a complex scene are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of waste sorting technology, and in particular to an intelligent management system for waste sorting and recycling based on the Internet of Things. Background Technology

[0002] Waste sorting stems from the environmental governance challenges brought about by accelerated global urbanization and population concentration. With industrialization and modernization, the amount of municipal solid waste generated has increased exponentially. In 2020, the total amount of municipal solid waste worldwide reached 2.2 billion tons, and China's annual municipal solid waste production exceeded 300 million tons. Traditional mixed collection and treatment methods result in a recyclable resource recovery rate of less than 20%, organic matter degradation produces large amounts of methane greenhouse gases, and inorganic components occupy more than 600,000 mu (approximately 40,000 hectares) of land for landfilling. Against this backdrop, the United Nations Environment Programme proposed the "Zero Waste Cities" initiative, and the European Union formulated the "Circular Economy Action Plan," aiming to achieve a municipal solid waste recycling rate of ≥35%.

[0003] In typical waste sorting and recycling, abnormal waste detection often relies on simple threshold judgments, lacking in-depth analysis that cross-validates pollution types, image features, and physical characteristics. This makes it difficult to accurately pinpoint the cause of anomalies, and the anomaly handling process depends on manual intervention, failing to achieve automated closed-loop management. Furthermore, based on fixed rules, it is difficult to dynamically adapt to the sorting standards of different cities or handle conflicts between multiple categories. Existing technologies also generally neglect end-to-end traceability capabilities, only recording the disposal timestamp without linking it to the disposal entity and equipment status, resulting in ambiguous traceability of abnormal waste and hindering accountability and system optimization.

[0004] To address the shortcomings of the existing technologies, this technical solution proposes an intelligent management system for waste sorting and recycling based on the Internet of Things. Summary of the Invention

[0005] This invention provides an intelligent management system for waste sorting and recycling based on the Internet of Things, in order to overcome the shortcomings of existing technologies.

[0006] On the one hand, the present invention provides an intelligent management system for waste sorting and recycling based on the Internet of Things, including: The data acquisition module is used to collect physical, image, and environmental information of the garbage in real time through a sensor network, and record the timestamp of the disposal. The feature extraction module is used to extract physical features, image features, and pollution features from physical information, image information, and environmental information. The waste type identification module is used to identify the type of waste disposed of based on image features, obtain an initial set of waste types, mark abnormal waste, and output abnormal tags; The waste sorting module is used to classify the waste based on the initial set of waste types and physical characteristics, and to generate a waste sorting strategy. The anomaly detection module is used to analyze the cause of anomalies based on anomaly markers, combined with image features and contamination features; The exception handling module is used to implement different processing schemes for abnormal garbage based on the cause of the exception and in combination with the preset abnormal garbage handling strategy, and output the abnormal garbage handling results. The anomaly tracing module is used to trace the source of abnormal waste based on waste disposal registration information and disposal timestamp, and to remind the waste disposal personnel based on the tracing results.

[0007] According to the IoT-based intelligent waste sorting and recycling management system provided by the present invention, the feature extraction module includes: a physical feature extraction unit, an image feature extraction unit, and an environmental feature extraction unit; the physical feature extraction unit is used to extract physical features from physical information, the image feature extraction unit is used to extract image features through machine learning methods, and the environmental feature extraction unit is used to extract pollution features present when waste is disposed of.

[0008] The intelligent waste sorting and recycling management system based on the Internet of Things provided by the present invention includes the following steps for extracting image features using machine learning methods: Extract historical garbage image data from publicly available garbage datasets; The historical garbage image data is standardized in size to output a standard-size image set; Construct a CNN model, including convolutional layers, activation layers, pooling layers, and fully connected layers; The CNN model is trained using a standard-sized image set, and the output is a pre-trained CNN model. By removing the fully connected layer, image information is input from the convolutional layer and output as first-level image features. The activation layer is used to introduce non-linear conditions, and the pooling layer is used to reduce the spatial dimension of the first-level image features according to the non-linear conditions, and output image features.

[0009] According to the IoT-based intelligent waste sorting and recycling management system provided by the present invention, the step of identifying the type of waste disposed of includes: Based on historical garbage image data, a garbage type identification model based on random forest is constructed, and the garbage type identification model is trained to obtain a pre-trained garbage type identification model. The image features are matched with the waste categories to output labeled image features; The labeled image features are input into a pre-trained waste type recognition model to identify waste types and output an initial set of waste types.

[0010] According to the IoT-based intelligent waste sorting and recycling management system provided by the present invention, the step of marking abnormal waste includes: Calculate the confidence level of the initial set of waste types and the confidence level value of each waste type. The confidence score is compared with a preset confidence threshold, as shown in the following formula:

[0011]

[0012] In the formula, P is the identification index, Si is the confidence level of garbage identification, and S0 is the preset confidence threshold; Let k be the Gaussian function, k be the Gaussian width, and x be the integration variable. The normalization coefficient is used. When P→1, the output is a normal garbage identification result; when P→0, the garbage identification result is marked as uncertain and an abnormal label is output.

[0013] According to the IoT-based intelligent waste sorting and recycling management system provided by the present invention, the steps for generating a waste sorting strategy include: Match the waste sorting system to the city it belongs to; Define the physical characteristics standards in the waste sorting system; Combining physical characteristic standards, the initial set of waste types is further classified according to physical characteristics and the waste classification system, and a more detailed set of waste types is output. Perform multi-category conflict determination on the set of subdivided waste types, and reclassify according to the preset priority strategy to output the waste classification strategy.

[0014] According to the IoT-based intelligent waste sorting and recycling management system provided by the present invention, the step of presetting the priority strategy includes: Based on safety priorities, the detailed waste categories are divided into hazardous waste and other waste categories. Based on safety priorities, other types of waste are divided into high-value waste, low-value waste, and abnormal waste. Abnormal waste is classified as other waste.

[0015] According to the IoT-based intelligent waste sorting and recycling management system provided by the present invention, the steps for analyzing the causes of anomalies include: Dynamically associate image features with contamination features through a cross-attention mechanism; The target detection model is used to locate the regions where contamination features exist in image information. Based on the location of the pollution, determine the type of pollution based on its characteristics.

[0016] According to the IoT-based intelligent waste sorting and recycling management system provided by the present invention, the step of locating the region where pollution features exist in image information using a target detection model includes: The lightweight YOLO model was chosen as the object detection model; Initialize the object detection model and load the pre-trained model weights; set the training parameters and train the object detection model using historical garbage image data, then output the optimal model weights. The target detection model is loaded with the optimal model weights. Image information is input into the target detection model for detection, and the bounding boxes of the areas where pollution features exist are output.

[0017] According to the IoT-based intelligent waste sorting and recycling management system provided by the present invention, the steps for tracing and tracking abnormal waste include: Combine the timestamp of the waste disposal to match the disposal time of the abnormal waste; Based on the time of disposal and registration information, the relevant entities associated with the abnormal waste are identified; If multiple deployment times are consecutive, the equipment fault tracing is triggered, and the start time of the time continuity is output. Based on the start time, export the device operation log and output the device fault detection results.

[0018] This invention provides an IoT-based intelligent management system for waste sorting and recycling. It employs a CNN model to extract image features and a random forest model to identify waste types, achieving deep fusion of multi-source heterogeneous data. By introducing a confidence assessment mechanism to dynamically verify classification results, it effectively reduces the false detection rate and significantly improves the accuracy and robustness of waste sorting in complex scenarios. Through a cross-attention mechanism to associate pollution features with image information, combined with a lightweight YOLO target detection model, it accurately locates polluted areas and achieves rapid diagnosis of anomalies. A safety priority-based hierarchical strategy and equipment fault tracing mechanism significantly reduce the cost of manual intervention, forming an intelligent closed loop from anomaly identification to problem repair. By adopting a collaborative classification strategy using physical and image features, combined with the dynamic matching capability of the urban waste sorting system, it can automatically adapt to the management regulations of different regions. Through multi-category conflict judgment and priority strategies, it optimizes resource allocation efficiency and improves the sorting rate and processing value of recyclables. Anomaly tracing can be accurate to the specific disposal entity and equipment node, supporting responsibility tracing and equipment operation and maintenance optimization, providing decision support for management departments, and promoting the institutionalization of waste sorting responsibility. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1This is a schematic diagram of the structure of the Internet of Things-based intelligent management system for waste sorting and recycling provided in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0022] Example 1: The following is combined Figure 1 This invention describes an IoT-based intelligent management system for waste sorting and recycling.

[0023] like Figure 1 As shown in the figure, the IoT-based intelligent waste sorting and recycling management system provided in this embodiment of the invention includes: a data acquisition module, a feature extraction module, a waste type identification module, a waste sorting module, an anomaly detection module, an anomaly handling module, and an anomaly tracing module. The system adopts an edge computing architecture, is deployed within an intelligent waste recycling station, and supports 5G communication protocol to achieve multi-device collaboration.

[0024] The data acquisition module is used to collect physical information (such as weight, volume, and material density), image information (RGB images and infrared thermal images from multispectral cameras), and environmental information (pollution information) of the disposed waste in real time through a sensor network, and records the disposal timestamp. The collected data is transmitted encrypted through a federated learning framework and supports offline resume and local caching mechanisms.

[0025] The feature extraction module is used to extract physical features (mass-to-volume ratio, material hardness), image features (texture, shape, color distribution), and pollution features (volatile organic compound concentration, particulate matter distribution) from physical information, image information, and environmental information. The feature extraction module includes: a physical feature extraction unit (integrating a strain sensor and pressure sensor array), an image feature extraction unit, and an environmental feature extraction unit (equipped with a miniature mass spectrometer and PM2.5 sensor). The physical feature extraction unit extracts physical features from the physical information, the image feature extraction unit extracts image features using machine learning methods, and the environmental feature extraction unit extracts pollution features present at the time of waste disposal. The steps for extracting image features using machine learning methods include: Historical waste image data was extracted from publicly available waste datasets. The dataset contains 2 million labeled images covering over 500 waste categories, and a comparative learning dataset was constructed using the MoCo-v3 algorithm.

[0026] Historical garbage image data is standardized in size, typically by bilinear interpolation to 640×640 resolution, to output a standard-sized image set.

[0027] When constructing the CNN model, the SE-ResNet module is introduced to enhance the channel attention mechanism, including convolutional layers (depth separable convolutional kernel size 3×3), Swish activation layers, ASPP pooling layers, and fully connected layers.

[0028] The CNN model is trained using a standard-sized image set, and the output of the pre-trained CNN model is trained using label smoothing techniques.

[0029] By removing fully connected layers, image information is input from convolutional layers and output as first-level image features. Activation layers are used to introduce nonlinear conditions, and pooling layers are used to reduce the spatial dimension of first-level image features according to the nonlinear conditions. Feature pyramid network (FPN) is used to fuse multi-scale features and output image features with spatial awareness.

[0030] The waste type identification module identifies the type of waste disposed of based on image features, obtains an initial set of waste types, marks abnormal waste, and outputs anomaly tags. Knowledge graphs can be introduced to assist classification, constructing a relational network encompassing material composition, physical properties, and environmental impact. The steps for identifying the type of waste disposed of include: Based on historical garbage image data, a garbage type identification model based on random forest was constructed and trained to obtain a pre-trained garbage type identification model. The XGBoost algorithm was used to optimize feature split points for each decision tree, and the Label Propagation algorithm was applied during training to address class imbalance.

[0031] Image features are matched with garbage categories to output labeled image features. A contrastive loss function is introduced during the labeling process to enhance the clustering of vector spaces for features of the same type.

[0032] The labeled image features are input into a pre-trained waste type recognition model to identify waste types and output an initial set of waste types. Simultaneously, a confidence matrix is ​​generated, recording the standard deviation of the predicted probability for each category.

[0033] Secondly, the steps for marking abnormal garbage include: The confidence level of the initial set of waste types is assessed, and the confidence intervals predicted by the Bayesian uncertainty estimation (BUE) model are quantified to calculate the confidence value of each waste type.

[0034] The confidence score is compared with a preset confidence threshold, as shown in the following formula:

[0035]

[0036] In the formula, P is the recognition index, and S... i S0 represents the confidence level for garbage identification, and S0 is the preset confidence threshold. Let be a Gaussian function, k be the Gaussian width, and x be the integration variable, existing only during the integration process. After integration, the result is independent of x and depends only on the upper limit of integration, S. i -S0 and the parameter, x represents only a mathematical tool used to define the form of the integrand. k The normalization coefficient is used. When P→1, a normal garbage identification result is output. When P→0, the garbage identification result is marked as uncertain, and an anomaly flag is output. The integration interval is (-∞, S). i The upper limit of -S0) determines the result: if S i -S0≥0, the integral covers the "body" of the Gaussian function, and the result is 1; otherwise, it only covers the left tail, and the result is 0.

[0037] The waste sorting module is used to classify waste based on an initial set of waste types and physical characteristics, generating a waste sorting strategy. The steps include: It matches the waste sorting system based on the user's city. It automatically matches local waste sorting standards using a built-in city database, supporting dynamic updates of the rule base for over 300 cities. By receiving the user's city information, it retrieves the corresponding city's sorting system (e.g., four-category / five-category) and loads the city's physical characteristic determination rules.

[0038] Define the physical characteristics standards in the waste sorting system. Characteristic dimensions include: material properties (metal / plastic / glass, etc.), morphological characteristics (solid / liquid / colloidal), and hazard labels (including heavy metals / corrosiveness / biohazards).

[0039] Combining physical characteristic standards, the initial set of waste types is further classified according to physical characteristics and the waste classification system, outputting a more detailed set of waste types. Feature weights can be set according to feature dimensions, specifically with hazard identification features having the highest weight, followed by material characteristics, and morphological characteristics having the basic weight.

[0040] The system performs multi-category conflict determination on the set of subdivided waste types, reclassifies the waste according to a preset priority strategy, and outputs the waste classification strategy. The steps for setting the preset priority strategy include: Based on safety priorities, the waste categories are divided into hazardous waste and other waste categories. The hazardous waste category can be further subdivided into: Level 1 Risk: Contains highly toxic substances (e.g., used batteries, expired medicines); Level 2 Risk: Possesses biological infectiousness (medical waste, animal carcasses); Level 3 Risk: Possesses physical hazards (sharp objects, glass shards). These substances must meet the requirements of categories HW01-HW49 in the National Hazardous Waste List. Other waste categories typically consist of residual waste that does not meet hazardous waste standards after testing by professional institutions.

[0041] Based on safety priorities, other categories of waste are divided into high-value waste, low-value waste, and unusual waste. The value is typically determined using a three-dimensional evaluation based on three aspects: material recycling value, processing cost, and market demand, which can be represented as:

[0042] Where β is the value index of waste in the three-dimensional evaluation, A is the unit recycling price, B is the sorting cost, C is the transportation cost, and α is the recycling rate of waste in the market. This can be obtained through historical survey data. For example: High-value waste: index ≥ 0.8 (e.g., PET plastic, aluminum). Low-value waste: 0.3 ≤ index < 0.8 (e.g., mixed paper, ceramic fragments). Abnormal waste: index < 0.3 or with value fluctuation risk. Data from commodity trading platforms can also be integrated to update the recycled resource price index monthly and establish an automatic sorting parameter adjustment system.

[0043] Abnormal waste is classified as "other waste." This typically includes cross-contamination waste, waste with variable composition, and waste with abnormal morphology. Cross-contamination waste is a mixture of different types of waste; it can be re-sorted through crushing and secondary sorting. Waste with variable composition is waste containing organic pollutants mixed with inorganic matter. Waste with abnormal morphology is usually waste that exceeds size limits; it can be crushed to reduce volume before sorting.

[0044] The anomaly detection module is used to analyze the causes of anomalies based on anomaly markers, combined with image features and contamination features. The steps include: Image features and contamination features are dynamically associated through a cross-attention mechanism. Specifically, a hierarchical alignment strategy is employed to achieve deep fusion of image and contamination features, including: First, a global association graph is constructed using the cross-attention mechanism to calculate the semantic similarity matrix between image features and contamination features. Second, a dynamic routing mechanism is introduced to establish local feature connections based on feature importance weights. Finally, a time-aware graph convolutional network is designed to model the diffusion path of contamination features in the image.

[0045] Furthermore, an adaptive feature enhancement pipeline is constructed, including: dynamically adjusting the receptive field of the convolutional kernel based on the saliency map of the contaminated region (adaptively switching from 3×3 to 7×7); strengthening key contaminated feature channels through learnable channel attention weights (weight range 0.5-1.5); and predicting motion trajectories of contaminated features in consecutive frames to compensate for the lack of temporal information.

[0046] The object detection model is used to locate the areas where contamination features exist in the image information.

[0047] Based on the region of existence, the pollution type of pollution features is determined. A cross-modal contrastive learning framework is established. Specifically, it involves feature pairs from different modalities of the same pollution event. An adversarial example generator is introduced to generate cross-modal interference pairs. InfoNCE loss combined with KL divergence constraints ensures feature distribution alignment. The role of InfoNCE loss is to learn robust cross-modal feature representations by maximizing the similarity of positive sample pairs and minimizing the similarity of negative sample pairs. It forces the latent spatial distribution of images and pollution features to align, making cross-modal features of the same pollution event closer and heterogeneous features farther apart. This addresses the semantic gap between modalities and improves the ability to distinguish pollution types. The calculation method is expressed as follows:

[0048] Among them, z i For image features corresponding to the same pollution time, c i Characteristics of pollution. i and c i This belongs to a positive sample pair. j Here, τ is the negative sample parameter, and τ is the temperature coefficient used to control the discrimination. j represents the negative sample index, where a negative sample is distinguished from the current positive sample z. i or c i Samples not belonging to the same pollution event. N i This represents the set of all negative samples.

[0049] Then, the supplementary classification action using KL divergence constraints is employed, including: The joint distribution p(z,c) of constrained image features and contamination features approximates the prior distribution p. prior (z,c), the formula is expressed as:

[0050] Among them, L kl Let be the KL divergence constraint loss, z be the image feature, and c be the contamination feature. p(z,c) is used to reflect the model's ability to model the correlation between input data features. prior (z,c) are typically designed as a standard Gaussian distribution or generated through adversarial training to provide a normalization objective for feature learning, constraining the model to avoid overfitting. klThis is the KL divergence calculation function, used to measure the difference between two parts, quantify the degree of deviation between the model feature distribution and the prior distribution, and force alignment.

[0051] Finally, two real-time collaborative mechanisms are used to divide the optimization objective into two parts, expressed by the following formula:

[0052]

[0053] Where γ is the contrastive learning loss weight, used to control L infoNCE The priority of L is set to a high value (e.g., 0.7) in the early stages of training to strengthen cross-modal alignment; it is reduced (e.g., 0.3) in the later stages to focus on distribution constraints. γ1 is the distribution constraint loss weight, used to control L KL The priority of L should be gradually increased during the later stages of training (e.g., from 0.3 to 0.7) to improve generalization ability. infoNCE To compare the learning loss, it is used to measure the difference in similarity between positive and negative sample pairs, ensuring that images of the same contamination event are closely associated with contamination features.

[0054] The steps described above for locating regions containing contamination features in an image using an object detection model include: The lightweight YOLOv8n model was chosen as the object detection model. YOLOv8n employs depthwise separable convolutions and model pruning techniques, with only about 3.2M parameters. It supports real-time inference (≥30FPS) on both CPUs and GPUs, making it suitable for deployment on edge devices. Based on an improved CSPNet Backbone network, computational redundancy is reduced through cross-stage partial connections; the Neck layer introduces an FPN+PAN structure to enhance multi-scale feature fusion capabilities and improve the accuracy of small object detection. Addressing the variable size of polluted areas (such as oil stains and plastic fragments) in industrial pollution detection, YOLOv8n's Anchor-Free mechanism can adaptively predict bounding boxes with different aspect ratios.

[0055] Initialize the object detection model and load the pre-trained model weights. Set the training parameters and train the object detection model using historical garbage image data, then output the optimal model weights.

[0056] The steps for initializing the object detection model include: annotating historical garbage images with bounding boxes of contaminated regions and category labels; randomly rotating, scaling, and shearing the historical garbage images to simulate various geometric transformations such as different shooting angles; then adjusting brightness and contrast using HSV space and adding Gaussian noise to enhance environmental robustness; and finally, splitting the training, validation, and test sets in an 8:1:1 ratio to ensure a balanced distribution of contaminated samples for each category.

[0057] Setting the training parameters specifically includes: presetting the image input size, typically 640×640. Configuring the optimizer (SGD + momentum (0.937) + weight decay (0.01)). Setting the learning rate using a cosine annealing strategy, with an initial value of 0.01 and 500 warm-up iterations.

[0058] When training the object detection model, if the validation set does not improve for 5 consecutive epochs, the training is terminated, and the 3 checkpoints with the highest F1 scores on the validation set are retained.

[0059] The optimal model weights are used to load the object detection model. Image information is input into the object detection model for detection, and the bounding boxes of the areas where pollution features exist are output. The input image is scaled to 640×640, and images with an aspect ratio > 2:1 are center-cropped and padded to avoid information loss. Low-confidence predictions are filtered before NMS (default 0.25) to balance the false negative and false positive rates. The IoU threshold is set to 0.45 to retain high-scoring boxes with low overlap. A linear decay strategy is used for overlapping boxes to improve the recall rate in densely polluted scenes.

[0060] The exception handling module is used to implement different processing schemes for abnormal garbage based on the cause of the exception and in combination with the preset exception garbage handling strategy, and output the exception garbage handling result.

[0061] The anomaly tracing module is used to trace the source of abnormal waste based on waste disposal registration information and disposal timestamps, and to remind waste disposal personnel based on the tracing results. The steps for tracing the source of abnormal waste include: Combine the timestamp of the waste disposal to match the disposal time of the abnormal waste.

[0062] Based on the disposal time and registration information, the associated entities of abnormal waste are identified. A three-tiered identification system can be established for multiple devices, including: Level 1: RFID tag ID; Level 2: User account system; Level 3: Device fingerprint ID. Blockchain-based notarization is implemented, with each identity associated with a 512-bit hash value.

[0063] If multiple deployment times are consecutive, the equipment fault tracing is triggered, and the start time of the consecutive time is output.

[0064] Based on the start time, export the device operation log and output the device fault detection results.

[0065] In summary, the IoT-based intelligent waste sorting and recycling management system provided by this invention achieves deep fusion of multi-source heterogeneous data by employing a CNN model to extract image features and a random forest model to identify waste types. By introducing a confidence assessment mechanism to dynamically verify classification results, the false detection rate is effectively reduced, significantly improving the accuracy and robustness of waste sorting in complex scenarios. A cross-attention mechanism is used to associate pollution features with image information, combined with a lightweight YOLO target detection model to accurately locate polluted areas, enabling rapid diagnosis of anomalies. A safety priority-based hierarchical strategy and equipment fault tracing mechanism significantly reduce the cost of manual intervention, forming an intelligent closed loop from anomaly identification to problem repair. By adopting a collaborative classification strategy using physical and image features, combined with the dynamic matching capability of the urban waste sorting system, it can automatically adapt to the management regulations of different regions. Through multi-category conflict judgment and priority strategies, resource allocation efficiency is optimized, improving the sorting rate and processing value of recyclables. Anomaly waste tracing can be accurate to the specific disposal entity and equipment node, supporting responsibility tracing and equipment operation and maintenance optimization, providing decision support for management departments, and promoting the institutionalization of waste sorting responsibility.

[0066] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent management system for waste sorting and recycling based on the Internet of Things, characterized in that, include: The data acquisition module is used to collect physical, image, and environmental information of the garbage in real time through a sensor network, and record the timestamp of the disposal. The feature extraction module is used to extract physical features, image features, and pollution features from the physical information, image information, and environmental information. The waste type identification module is used to identify the type of waste disposed of based on the image features, obtain an initial waste type set, mark abnormal waste, and output an abnormal label; The waste sorting module is used to sort the waste based on the initial set of waste types and the physical characteristics, and to generate a waste sorting strategy. An anomaly detection module is used to analyze the cause of the anomaly based on the anomaly marker, combined with the image features and the contamination features; An exception handling module is used to implement different processing schemes for the abnormal waste according to the cause of the exception and in combination with a preset abnormal waste processing strategy, and output the abnormal waste processing result. The anomaly tracing module is used to trace the abnormal waste based on the waste disposal registration information and the disposal timestamp, and to remind the waste disposal person based on the tracing results.

2. The IoT-based intelligent management system for waste sorting and recycling as described in claim 1, characterized in that, The feature extraction module includes a physical feature extraction unit, an image feature extraction unit, and an environmental feature extraction unit; the physical feature extraction unit is used to extract physical features from the physical information, the image feature extraction unit is used to extract image features using machine learning methods, and the environmental feature extraction unit is used to extract pollution features present during waste disposal.

3. The IoT-based intelligent management system for waste sorting and recycling according to claim 2, characterized in that, The steps for extracting image features using machine learning methods include: Extract historical garbage image data from publicly available garbage datasets; The historical waste image data is standardized in size to output a standard-size image set. Construct a CNN model, including convolutional layers, activation layers, pooling layers, and fully connected layers; The CNN model is trained using the standard-sized image set to output a pre-trained CNN model. Remove the fully connected layer, input the image information from the convolutional layer and output first-level image features, the activation layer is used to introduce non-linear conditions, the pooling layer is used to reduce the spatial dimension of the first-level image features according to the non-linear conditions, and output the image features.

4. The intelligent waste sorting and recycling management system based on the Internet of Things according to claim 1, characterized in that, The steps for identifying the type of waste disposed of include: Based on historical garbage image data, a garbage type identification model based on random forest is constructed, and the garbage type identification model is trained to obtain a pre-trained garbage type identification model. The image features are matched with waste categories to output labeled image features; The labeled image features are input into the pre-trained waste type recognition model to identify waste types and output an initial waste type set.

5. The IoT-based intelligent management system for waste sorting and recycling according to claim 4, characterized in that, The steps for marking abnormal garbage include: A confidence assessment is performed on the initial set of waste types, and a confidence value for each waste type is calculated. The confidence value is compared with a preset confidence threshold, as expressed by the following formula: ； ； In the formula, P is the identification index, Si is the confidence level of garbage identification, and S0 is the preset confidence threshold; Let k be the Gaussian function, k be the Gaussian width, and x be the integration variable. The normalization coefficient is used. When P→1, the output is a normal garbage identification result; when P→0, the garbage identification result is marked as uncertain and an abnormal label is output.

6. The IoT-based intelligent management system for waste sorting and recycling according to claim 5, characterized in that, The steps for generating the waste sorting strategy include: Match the waste sorting system to the city it belongs to; Define the physical characteristic standards in the aforementioned waste sorting system; Based on the physical characteristics criteria, the initial set of waste types is further classified according to the physical characteristics and the waste classification system, and a more detailed set of waste types is output. The system performs multi-category conflict determination on the set of subdivided waste types, reclassifies them according to a preset priority strategy, and outputs the waste classification strategy.

7. The IoT-based intelligent management system for waste sorting and recycling according to claim 6, characterized in that, The steps for setting the priority policy include: Based on safety priorities, the detailed set of waste categories is divided into a hazardous waste category and a other waste category category. Based on the aforementioned security priority, the other categories of waste are divided into high-value waste, low-value waste, and abnormal waste. The abnormal waste is classified as other waste.

8. The IoT-based intelligent management system for waste sorting and recycling according to claim 7, characterized in that, The steps for analyzing the causes of anomalies include: The image features and the pollution features are dynamically associated through a cross-attention mechanism; The presence area of ​​the pollution feature in the image information is located using a target detection model; Based on the area where the pollution exists, determine the type of pollution characterized by the pollution.

9. The IoT-based intelligent management system for waste sorting and recycling according to claim 8, characterized in that, The steps of locating the region where the contamination feature exists in the image information using an object detection model include: The lightweight YOLO model was selected as the target detection model. Initialize the target detection model, load the pre-trained model weights; set the training parameters, and train the target detection model using historical garbage image data, outputting the optimal model weights; The target detection model is loaded using the optimal model weights, the image information is input into the target detection model for detection, and the bounding box of the area where the pollution feature exists is output.

10. The intelligent waste sorting and recycling management system based on the Internet of Things according to claim 1, characterized in that, The steps for tracing the source of the abnormal waste include: By combining the disposal timestamp, the disposal time of the abnormal waste is matched; Based on the disposal time and the registration information, the associated entity of the abnormal waste is determined; If multiple deployment times show a temporal continuity, then equipment fault tracing is triggered, and the start time of the temporal continuity is output. Based on the stated start time, export the device operation log and output the device fault detection results.

Images