An intelligent supervision method and system for solid waste recycling
Through deep learning models and random forest algorithms, the intelligent recycling system is optimized, and the problems of low efficiency and poor accuracy in classification and recycling of electronic waste and hazardous waste are solved, and more efficient waste management is achieved.
Patent Information
- Application Number
- CN202410457511.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-16
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-04-16
AI Technical Summary
Intelligent recycling systems are inefficient and poorly accurate in the classification and recycling of electronic waste and hazardous waste.
Deep learning model combined with random forest algorithm is adopted to clean, normalize, reduce dimensionality and optimize by obtaining waste type and state data, output recycling value and processing priority, and improve the classification and recycling process.
Improve classification efficiency and recycling accuracy, ensuring the correct classification and treatment of electronic waste and hazardous waste.
Smart Images

Figure CN118195598B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of waste recycling technology, and in particular to an intelligent supervision method and system for solid waste recycling. Background Art
[0002] With the rapid development of industrialization and urbanization, the amount of solid waste generated is also increasing. This trend has prompted relevant industries and government departments to take more effective recycling and resource utilization measures to reduce the impact on the environment and promote sustainable development. In terms of technology, the application of intelligent supervision methods is gradually increasing, including the use of Internet of Things technology to improve recycling efficiency and supervision capabilities.
[0003] Among the relevant technical means, smart recycling bins and smart scales can automatically measure and record the weight of recyclables, and connect to the network to monitor the recycling situation in real time. This not only improves the recycling efficiency, but also encourages the public to participate in recycling activities through a points incentive program.
[0004] Regarding the above technical solution, although the weight of recyclables can be automatically measured and recorded through smart recycling bins and smart scales, the smart recycling system has problems with low classification efficiency and poor recycling accuracy when classifying and recycling some specific wastes, such as electronic waste and hazardous waste. Summary of the invention
[0005] In order to improve the problems of low classification efficiency and poor recycling accuracy in the classification and recycling of electronic waste and hazardous waste, the present application provides an intelligent supervision method and system for solid waste recycling.
[0006] The present invention provides an intelligent supervision method for solid waste recycling, comprising the following steps: obtaining type data and state data of waste, inputting the type data and the state data into a preset waste identification model to obtain identification data, classifying the identification data through a classifier to distinguish normal data from noise data; obtaining cleaning rules corresponding to specific attributes of the type data and the state data, and cleaning the normal data and the noise data according to the cleaning rules; integrating the cleaned normal data and the noise data into a unified format for normalization, encoding and converting the normalized data to obtain a data set, performing high-dimensional data mapping on the data set to optimize the position of the data points of the data set in a low-dimensional space, maintaining the relative distance between the data points in the high-dimensional space, and obtaining a reduced-dimensional data set; inputting the reduced-dimensional data set into a pre-trained random forest algorithm, outputting the recovery value and processing priority of the waste, and optimizing the waste identification model according to the recovery value and the processing priority; integrating the optimized waste identification model into a system of a display interface, wherein the display interface is used to monitor and adjust the classification and recycling process of the waste in real time.
[0007] As a preferred solution, the type data includes the physical composition, weight, volume and shape of the waste; the status data includes the freshness, contamination degree and decomposition state of the waste.
[0008] As a preferred solution, the waste recognition model is constructed based on a deep learning model of the FasterR-CNN architecture and initialized using parameters pre-trained on the ImageNet database; the region proposal network in the FasterR-CNN is trained separately, and the underlying weights of the feature extraction network in the FasterR-CNN are frozen; the feature extraction network is fine-tuned using a waste data set; wherein the waste data set contains type data and status data of a variety of wastes; the trained region proposal network and the fine-tuned feature extraction network share the same set of convolutional layers; wherein the convolutional layers are used to extract image features; and the region proposal network is used to propose candidate images based on the image features. An object bounding box is selected, and a candidate region is confirmed according to the candidate object bounding box; the candidate region is input into a RoIPooling layer, and the RoIPooling layer standardizes the size of each candidate region; the standardized candidate region is input into a fully connected layer of a feature extraction network, and the fully connected layer classifies each candidate region and adjusts the candidate object bounding box; the category of each candidate region and the coordinates of the adjusted candidate object bounding box are output, so as to obtain the waste recognition model; when the type data and the state data are input into the waste recognition model, the waste recognition model maps the type data and the state data to a specific waste category, and outputs recognition data.
[0009] As a preferred embodiment, the step of obtaining cleaning rules corresponding to specific attributes of the type data and the status data includes: presetting dynamic cleaning rules according to the physical composition, weight, volume and shape of the type data, and the freshness, degree of contamination and decomposition state of the status data; wherein the dynamic cleaning rules use machine learning algorithms to automatically adjust data cleaning parameters according to the specific circumstances of the waste.
[0010] As a preferred solution, before the step of performing high-dimensional data mapping on the data set to obtain a data set after dimensionality reduction, it includes: cross-validating the data set through a support vector machine to verify the accuracy of the normal data and the noise data after cleaning, and adjusting the cleaning rules according to the verification results.
[0011] As a preferred solution, the step of performing high-dimensional data mapping on the data set to optimize the positions of the data points of the data set in the low-dimensional space, maintain the relative distances between the data points in the high-dimensional space, and obtain the data set after dimensionality reduction includes: performing high-dimensional data mapping on the data set by t-distributed random neighbor embedding technology to optimize the positions of the data points of the data set in the low-dimensional space, maintain the relative distances between the data points in the high-dimensional space, and obtain the data set after dimensionality reduction; adjusting the perplexity and learning rate of the t-SNE algorithm according to the physical composition, weight, volume, shape, freshness, degree of contamination and decomposition state characteristics in the data set after dimensionality reduction.
[0012] As a preferred solution, the random forest algorithm training process is as follows: construct multiple decision trees through the characteristics of the physical composition, weight, volume, shape, freshness, degree of contamination and decomposition state of the reduced dimension data set; use different waste data subsets and feature subsets to train each decision tree separately; wherein the waste data subset includes waste samples collected from different regions; the feature subset includes the most representative features screened by the feature selection algorithm; integrate the outputs of all decision trees, evaluate the recycling value and treatment priority of the waste, and adjust the weights of the decision trees according to real-time market data and environmental policies.
[0013] As a preferred solution, the step of optimizing the waste identification model according to the recycling value and the treatment priority includes: evaluating the accuracy of the waste identification model in predicting the recycling value and treatment priority of the waste according to the historical data of waste classification and recycling through a feedback loop mechanism; if the performance of the waste identification model in predicting the recycling value and treatment priority does not meet the predetermined standards, optimizing the parameters of the waste identification model through a genetic algorithm, and then applying the optimized parameters to the waste identification model.
[0014] As a preferred solution, before the step of integrating the optimized waste identification model into the system display interface, it includes: presetting a Web-based user interface, which is used to display the real-time classification and recycling status of waste and adjust the threshold and weight of the waste identification model.
[0015] The present application also provides an intelligent supervision system for solid waste recycling, including: a data acquisition unit, used to acquire type data and status data of waste, input the type data and the status data into a preset waste recognition model to obtain recognition data, classify the recognition data through a classifier, and distinguish normal data from noise data; a data cleaning unit, used to acquire cleaning rules corresponding to specific attributes of the type data and the status data, and clean the normal data and the noise data according to the cleaning rules; a data integration unit, used to integrate the cleaned normal data and the noise data into a unified format for normalization, and normalize the normalized data The data is encoded and converted to obtain a data set, and the data set is subjected to high-dimensional data mapping to optimize the position of the data points of the data set in the low-dimensional space, maintain the relative distance between the data points in the high-dimensional space, and obtain the reduced-dimensional data set; a model optimization unit is used to input the reduced-dimensional data set into a pre-trained random forest algorithm, output the recycling value and treatment priority of the waste, and optimize the waste identification model according to the recycling value and the treatment priority; a model integration unit is used to integrate the optimized waste identification model into a display interface system, and the display interface is used to monitor and adjust the waste classification and recycling process in real time.
[0016] Compared with the prior art, the present application has the following beneficial effects: high classification efficiency and high recycling accuracy. The type data and status data of the waste are obtained through sensors, and these data are input into a preset waste identification model. The waste identification model uses a classifier to process the data and distinguish normal data from noise data; next, cleaning rules are formulated according to the specific attributes of the type data and status data, and these rules are applied to clean the data, and then the cleaned data are integrated and normalized, and then encoded and converted, and the data set is optimized through high-dimensional data mapping technology to maintain the relative distance between data points in a low-dimensional space to obtain a reduced-dimensional data set, which is input into a random forest algorithm, which outputs the recycling value and processing priority of the waste, and optimizes the waste identification model accordingly; finally, the optimized waste identification model is integrated into a system of a display interface, which is used to monitor and adjust the waste classification and recycling process in real time, thereby improving the problem of low classification efficiency and poor recycling accuracy when electronic waste and hazardous waste are classified and recycled. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0018] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.
[0019] Figure 1 It is a flow chart of an intelligent supervision method for solid waste recycling provided by an embodiment of the present invention;
[0020] Figure 2 It is a schematic block diagram of the structure of the intelligent supervision system provided by an embodiment of the present invention.
[0021] Description of reference numerals:
[0022] 10. Intelligent supervision system; 11. Data acquisition unit; 12. Data cleaning unit; 13. Data integration unit; 14. Model optimization unit; 15. Model integration unit. DETAILED DESCRIPTION
[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0024] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.
[0025] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.
[0026] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.
[0027] The technical solution of the present invention is further described below with reference to the accompanying drawings and through specific implementation methods.
[0028] Embodiment 1:
[0029] like Figure 1 As shown, the intelligent supervision method for solid waste recycling provided in the embodiment of the present application includes steps S100 to S500.
[0030] Step S100, obtaining type data and status data of waste, inputting the type data and status data into a preset waste identification model to obtain identification data, and classifying the identification data through a classifier to distinguish normal data from noise data.
[0031] In this step, the type data and status data of the waste are obtained through sensors. The type data mainly includes the physical composition, weight, volume and shape of the waste; the status data mainly includes the freshness, degree of contamination and decomposition state of the waste, which is collected by weight sensors, volume measuring devices, material analyzers and other equipment. The type data and status data are processed by deep learning algorithms, features are extracted, and input into the waste identification model; the identification data is then classified by a classifier into normal data that is recyclable and non-recyclable, thereby improving the accuracy and efficiency of waste classification.
[0032] Step S200, obtaining cleaning rules corresponding to specific attributes of type data and state data, and cleaning normal data and noise data according to the cleaning rules.
[0033] In this step, cleaning rules are formulated based on the specific attributes of the type data and status data of the waste. For example, for electronic waste, the cleaning rules will consider the element composition and battery status of its circuit boards; these cleaning rules are applied to clean normal data and noise data through data cleaning algorithms such as outlier detection and data correction techniques.
[0034] Step S300, integrate the cleaned normal data and noise data into a unified format for normalization, encode and convert the normalized data to obtain a data set, and perform high-dimensional data mapping on the data set to optimize the position of the data points of the data set in the low-dimensional space, maintain the relative distance between the data points in the high-dimensional space, and obtain the reduced-dimensional data set.
[0035] In this step, the cleaned normal data and noise data are integrated into a unified format and normalized. For example, all weight data are converted to kilograms and volume data are converted to cubic meters. Then, the normalized data is encoded and converted, such as using one-hot encoding or label encoding. The data set is reduced in dimension through high-dimensional data mapping technology, such as t-distributed random neighbor embedding technology and kernel principal component analysis, to optimize the distribution of data points in low-dimensional space, making the data more suitable for processing by machine learning models and improving the performance of the model.
[0036] Step S400, input the reduced dimension data set into a pre-trained random forest algorithm, output the recycling value and treatment priority of the waste, and optimize the waste identification model according to the recycling value and treatment priority.
[0037] In this step, the reduced-dimensional dataset is input into a pre-trained random forest algorithm, which outputs the recycling value and treatment priority of the waste based on the dataset. For example, metal and glass are assigned higher recycling values, while organic waste is prioritized based on its decomposition speed, thereby improving the economic and environmental benefits of waste treatment.
[0038] Step S500, integrating the optimized waste identification model into a system display interface, where the display interface is used to monitor and adjust the waste classification and recycling process in real time.
[0039] In this step, the optimized waste identification model is integrated into the system of the display interface. The display interface provides an intuitive user interface that displays the waste classification results, recycling value and processing priority. The system can also monitor and adjust the waste classification and recycling process in real time, such as automatically adjusting the conveyor belt speed and the allocation of recycling bins, thereby improving the intelligence level of the waste management system and making it more efficient and user-friendly.
[0040] In the above embodiment, the sensor is responsible for collecting the type data and state data of the waste, providing a detailed feature description for the waste, and then using the data collected by the sensor to train and optimize the waste identification model, so that the waste identification model can learn the characteristics of the waste and accurately classify it, and ensure the quality of the data through the data cleaning process, improve the accuracy of the data by removing noise and outliers, and the cleaned data is more suitable for the processing of the machine learning model, thereby improving the performance of the model. High-dimensional data mapping is used to optimize the position of the data points of the data set in the low-dimensional space, maintain the relative distance between the data points in the high-dimensional space, and reduce the dimension of the data while retaining the most important features, which helps to improve the efficiency of data processing and reduce the consumption of computing resources. Finally, the recycling value and processing priority of the waste are evaluated by the random forest algorithm. The output of the random forest algorithm can be used to guide the classification and recycling of waste, as well as optimize resource allocation and processing processes, thereby improving the problems of low classification efficiency and poor recycling accuracy when classifying and recycling electronic waste and hazardous waste.
[0041] Embodiment 2:
[0042] In step S100, the waste recognition model applied is a deep learning model based on the FasterR-CNN architecture, which is initialized by using the pre-trained parameters on the ImageNet database; the region proposal network in FasterR-CNN is trained separately, and the underlying weights of the feature extraction network in FasterR-CNN are frozen; the feature extraction network is fine-tuned using a waste data set; wherein the waste data set contains type data and status data of a variety of wastes; the trained region proposal network and the fine-tuned feature extraction network share the same set of convolutional layers; wherein the convolutional layers are used to extract image features; the region proposal network is used to extract image features; the feature extraction ... The network proposes a candidate object bounding box based on the image features, and confirms the candidate area based on the candidate object bounding box; the candidate area is input into the RoIPooling layer, and the RoIPooling layer standardizes the size of each candidate area; the standardized candidate area is input into the fully connected layer of the feature extraction network, and the fully connected layer classifies each candidate area and adjusts the candidate object bounding box; the category of each candidate area and the coordinates of the adjusted candidate object bounding box are output to obtain a waste recognition model; when the type data and the status data are input into the waste recognition model, the waste recognition model maps the type data and the status data to a specific waste category and outputs the recognition data.
[0043] By applying a deep learning model based on the FasterR-CNN architecture, the system is able to leverage the power of its region proposal network and feature extraction network to efficiently identify and classify waste images. For example, when the system receives an image containing multiple types of waste, it is able to accurately identify each object, such as a plastic bottle, paper, or metal can, and assign them to the corresponding recycling category. This precise classification not only improves the efficiency of the recycling process, but also ensures the correct use and treatment of resources.
[0044] In step S200, cleaning rules corresponding to specific attributes of type data and status data are obtained, including: presetting dynamic cleaning rules according to the physical composition, weight, volume and shape of the type data, and the freshness, degree of contamination and decomposition state of the status data; wherein the dynamic cleaning rules use machine learning algorithms to automatically adjust the parameters of data cleaning according to the specific circumstances of the waste.
[0045] The application of dynamic cleaning rules enables the system to automatically adjust data cleaning parameters according to the specific conditions of the waste, thereby improving data quality. For example, if a waste sample is difficult to identify due to oil pollution, the system can adjust the cleaning rules to remove these noise data and ensure the correct classification of the waste.
[0046] In step S300, before performing high-dimensional data mapping on the data set to obtain a reduced-dimensional data set, the method further includes: cross-validating the data set through a support vector machine to verify the accuracy of the cleaned normal data and noise data, and adjusting the cleaning rules according to the verification results.
[0047] Cross-validation of support vector machines helps the system evaluate and optimize data cleaning rules to ensure the accuracy of the data set. Even when there are outliers or deviations in the data set, the system can correct these problems by adjusting the cleaning rules, thereby providing more accurate data for subsequent steps.
[0048] At the same time, the step of performing high-dimensional data mapping on the data set to optimize the positions of the data points of the data set in the low-dimensional space, maintain the relative distance between the data points in the high-dimensional space, and obtain the data set after dimensionality reduction includes: performing high-dimensional data mapping on the data set through t-distributed random neighbor embedding technology to optimize the positions of the data points of the data set in the low-dimensional space, maintain the relative distance between the data points in the high-dimensional space, and obtain the data set after dimensionality reduction; adjusting the perplexity and learning rate of the t-SNE algorithm according to the physical composition, weight, volume, shape, freshness, degree of contamination and decomposition state characteristics of the data set after dimensionality reduction.
[0049] In the process of high-dimensional data mapping, the application of t-SNE technology allows the system to better distinguish and identify the characteristics of waste in the reduced-dimensional data set. For example, by adjusting the perplexity and learning rate of the t-SNE algorithm, the system can more clearly distinguish different types of plastics, even if their characteristics in high-dimensional space are very similar.
[0050] In step S400, the training process of the applied random forest algorithm is as follows: multiple decision trees are constructed through the characteristics of physical composition, weight, volume, shape, freshness, degree of contamination and decomposition state of the reduced dimension data set; each decision tree is trained separately using different waste data subsets and feature subsets; wherein the waste data subset includes waste samples collected from different regions; the feature subset includes the most representative features screened by the feature selection algorithm; the outputs of all decision trees are integrated to evaluate the recycling value and treatment priority of the waste, and the weights of the decision trees are adjusted according to real-time market data and environmental policies.
[0051] The use of random forest algorithms provides a powerful tool for assessing the recycling value and treatment priority of waste. By building multiple decision trees and integrating their outputs, the system is able to analyze waste from multiple perspectives, providing a more comprehensive assessment that helps optimize the waste treatment process. For example, the system can determine which waste should be recycled first and which can be delayed.
[0052] At the same time, when optimizing the waste identification model according to the recycling value and treatment priority, it also includes: evaluating the accuracy of the waste identification model in predicting the recycling value and treatment priority of the waste based on the historical data of waste classification and recycling through a feedback loop mechanism; if the performance of the waste identification model in predicting the recycling value and treatment priority does not meet the predetermined standards, optimizing the parameters of the waste identification model through a genetic algorithm, and then applying the optimized parameters to the waste identification model.
[0053] Through the combination of feedback loop mechanisms and genetic algorithms, the waste identification model is able to self-learn and adapt, thereby continuously improving its performance in predicting the recycling value and treatment priority of waste. For example, if the model performs poorly in predicting the recycling value of waste of a specific material, such as electronic waste, the genetic algorithm can optimize the model parameters, such as adjusting weights and biases, to improve the accuracy of the prediction by simulating the process of natural selection. The optimization process can be based on historical data to ensure that the model can adapt to changes in different types of waste and fluctuations in market value.
[0054] In step S500, before the optimized waste identification model is integrated into the system of the display interface, the method further includes: presetting a web-based user interface, the user interface being used to display the real-time classification and recycling status of waste and to adjust the threshold and weight of the waste identification model.
[0055] The integration of the optimized waste identification model with a web-based user interface provides operators with an intuitive platform for monitoring and adjusting the waste sorting and recycling process. This integration improves the transparency and operability of the system, allowing operators to make quick decisions based on real-time data, such as adjusting sorting thresholds or reallocating resources.
[0056] Example 3
[0057] like Figure 2 As shown, the present application also provides an intelligent monitoring system 10 for solid waste recycling, including a data acquisition unit 11, a data cleaning unit 12, a data integration unit 13, a model optimization unit 14 and a model integration unit 15; wherein the data acquisition unit 11 is mainly used to obtain the type data and status data of the waste, input the type data and status data into a preset waste identification model, obtain identification data, and classify the identification data through a classifier to distinguish normal data from noise data. By utilizing advanced sensor technology and deep learning models such as FasterR-CNN, waste can be accurately identified and classified. For example, the system can distinguish between organic waste and inorganic waste to ensure that they are correctly allocated to the corresponding recycling process.
[0058] The data cleaning unit 12 is mainly used to obtain cleaning rules corresponding to specific attributes of type data and state data, and clean normal data and noise data according to the cleaning rules. By applying dynamic cleaning rules, the cleaning parameters are automatically adjusted to adapt to the characteristics of different wastes, thereby improving data quality. This means that even if the state of the waste changes, such as label residues on plastic bottles or corruption of food waste, the system can still accurately process the data.
[0059] The data integration unit 13 is mainly used to integrate the cleaned normal data and noise data into a unified format for normalization, perform encoding conversion on the normalized data to obtain a data set, and perform high-dimensional data mapping on the data set to optimize the position of the data points of the data set in the low-dimensional space, maintain the relative distance between the data points in the high-dimensional space, and obtain a reduced-dimensional data set. The cleaned data is normalized and encoded, and then high-dimensional data mapping techniques such as t-SNE are applied to optimize the representation of the data in the low-dimensional space. This helps to improve the accuracy of the classification and recovery value prediction of the random forest algorithm in the model optimization unit 14.
[0060] The model optimization unit 14 is mainly used to input the reduced-dimensional data set into the pre-trained random forest algorithm, output the recycling value and treatment priority of the waste, and optimize the waste identification model according to the recycling value and treatment priority. The recycling value and treatment priority of the waste are evaluated by using the random forest algorithm, and the waste identification model is optimized according to these evaluation results. In this way, the system can adjust its prediction according to the real-time market value and environmental policy to ensure that the waste treatment is more economical and efficient.
[0061] The model integration unit 15 is mainly used to integrate the optimized waste identification model into the system of the display interface, and the display interface is used to monitor and adjust the waste classification and recycling process in real time. By integrating the optimized waste identification model into the system of the display interface, a user-friendly interface is provided for real-time monitoring of the waste classification and recycling status. The operator can adjust the model parameters through this interface to cope with different operating conditions and requirements.
[0062] It should be noted that technical personnel in the relevant technical field can clearly understand that, for the convenience and conciseness of description, the specific working process of the system and each unit described above can refer to the corresponding process in the aforementioned embodiment of the intelligent supervision method for solid waste recycling, and will not be repeated here.
[0063] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. An intelligent supervision method for solid waste recycling, It is characterized in that include: Obtaining type data and status data of waste, wherein the type data includes the physical composition, weight, volume and shape of the waste; the status data includes the freshness, degree of contamination and decomposition state of the waste; inputting the type data and the status data into a preset waste recognition model, wherein the waste recognition model is constructed based on a deep learning model of the FasterR-CNN architecture and initialized using parameters pre-trained on the ImageNet database; separately training the region proposal network in the FasterR-CNN and freezing the underlying weights of the feature extraction network in the FasterR-CNN; fine-tuning the feature extraction network using a waste data set; wherein the waste data set contains type data and status data of multiple types of waste; allowing the trained region proposal network and the fine-tuned feature extraction network to share the same set of convolutional layers; wherein the convolutional layers Used to extract image features; propose a candidate object bounding box according to the image features through the region proposal network, and confirm the candidate region according to the candidate object bounding box; input the candidate region into the RoIPooling layer, and the RoIPooling layer standardizes the size of each candidate region; input the standardized candidate region into the fully connected layer of the feature extraction network, and the fully connected layer classifies each candidate region and adjusts the candidate object bounding box; output the category of each candidate region and the coordinates of the adjusted candidate object bounding box, so as to obtain the waste recognition model; when the type data and the state data are input into the waste recognition model, the waste recognition model maps the type data and the state data to a specific waste category, and outputs recognition data, and classifies the recognition data through a classifier to distinguish normal data from noise data; Acquire cleaning rules corresponding to specific attributes of the type data and the state data, and clean the normal data and the noise data according to the cleaning rules; Integrate the cleaned normal data and the noise data into a unified format for normalization, perform encoding conversion on the normalized data to obtain a data set, perform high-dimensional data mapping on the data set to optimize the position of the data points of the data set in the low-dimensional space, maintain the relative distance between the data points in the high-dimensional space, and obtain a reduced-dimensional data set; Inputting the dimension-reduced data set into a pre-trained random forest algorithm, outputting the recycling value and treatment priority of the waste, and evaluating the accuracy of the waste identification model in predicting the recycling value and treatment priority of the waste based on the historical data of waste classification and recycling through a feedback loop mechanism; if the performance of the waste identification model in predicting the recycling value and treatment priority does not meet the predetermined standards, optimizing the parameters of the waste identification model through a genetic algorithm, and then applying the optimized parameters to the waste identification model; The optimized waste identification model is integrated into a system of a display interface, wherein the display interface is used to monitor and adjust the waste classification and recycling process in real time.
2. The intelligent supervision method for solid waste recycling according to claim 1, It is characterized in that The step of obtaining cleaning rules corresponding to specific attributes of the type data and the state data includes: Dynamic cleaning rules are preset according to the physical composition, weight, volume and shape of the type data, and the freshness, contamination degree and decomposition state of the status data; wherein the dynamic cleaning rules use machine learning algorithms to automatically adjust the parameters of data cleaning according to the specific circumstances of the waste.
3. The intelligent supervision method for solid waste recycling according to claim 1, It is characterized in that Before the step of performing high-dimensional data mapping on the data set to obtain a data set after dimension reduction, the method includes: The data set is cross-validated by a support vector machine to verify the accuracy of the normal data and the noise data after cleaning, and the cleaning rules are adjusted according to the verification results.
4. The intelligent supervision method for solid waste recycling according to claim 1, It is characterized in that The step of performing high-dimensional data mapping on the data set to optimize the positions of the data points of the data set in the low-dimensional space and maintain the relative distances between the data points in the high-dimensional space to obtain a data set after dimensionality reduction includes: Performing high-dimensional data mapping on the data set by using t-distributed random neighbor embedding technology to optimize the positions of data points of the data set in the low-dimensional space, maintain the relative distances between data points in the high-dimensional space, and obtain a data set after dimensionality reduction; The perplexity and learning rate of the t-SNE algorithm are adjusted according to the physical composition, weight, volume, shape, freshness, contamination degree and decomposition state characteristics in the reduced dimension data set.
5. The intelligent supervision method for solid waste recycling according to claim 1, It is characterized in that The random forest algorithm training process is as follows: Constructing multiple decision trees based on the characteristics of the physical composition, weight, volume, shape, freshness, degree of contamination and decomposition state of the reduced-dimensional data set; Using different waste data subsets and feature subsets, each decision tree is trained separately; wherein the waste data subset includes waste samples collected from different regions; and the feature subset includes the most representative features selected by a feature selection algorithm; Integrate the outputs of all decision trees to evaluate the recycling value and treatment priority of waste, and adjust the weights of decision trees based on real-time market data and environmental policies.
6. The intelligent monitoring method for solid waste recycling according to claim 1, It is characterized in that Before the step of integrating the optimized waste identification model into the system of the display interface, the method includes: A web-based user interface is preset, and the user interface is used to display the real-time classification and recycling status of waste and adjust the threshold and weight of the waste identification model.
7. An intelligent monitoring system for solid waste recycling, It is characterized in that include: A data acquisition unit is provided for acquiring type data and status data of waste, wherein the type data includes physical composition, weight, volume and shape of the waste; the status data includes freshness, degree of contamination and decomposition state of the waste; the type data and the status data are input into a preset waste recognition model, wherein the waste recognition model is constructed based on a deep learning model of the FasterR-CNN architecture and initialized using parameters pre-trained on the ImageNet database; the region proposal network in the FasterR-CNN is trained separately, and the underlying weights of the feature extraction network in the FasterR-CNN are frozen at the same time; the feature extraction network is fine-tuned using a waste data set; wherein the waste data set includes type data and status data of multiple types of waste; the trained region proposal network and the fine-tuned feature extraction network share the same set of convolutional layers; wherein the The convolution layer is used to extract image features; the region proposal network proposes a candidate object bounding box according to the image features, and confirms the candidate region according to the candidate object bounding box; the candidate region is input into the RoIPooling layer, and the RoIPooling layer standardizes the size of each candidate region; the standardized candidate region is input into the fully connected layer of the feature extraction network, and the fully connected layer classifies each candidate region and adjusts the candidate object bounding box; the category of each candidate region and the coordinates of the adjusted candidate object bounding box are output, so as to obtain the waste recognition model; when the type data and the state data are input into the waste recognition model, the waste recognition model maps the type data and the state data to a specific waste category, and outputs recognition data, and the recognition data is classified by the classifier to distinguish normal data from noise data; A data cleaning unit, configured to obtain cleaning rules corresponding to specific attributes of the type data and the state data, and clean the normal data and the noise data according to the cleaning rules; A data integration unit is used to integrate the cleaned normal data and the noise data into a unified format for normalization, perform encoding conversion on the normalized data to obtain a data set, perform high-dimensional data mapping on the data set to optimize the position of the data points of the data set in the low-dimensional space, maintain the relative distance between the data points in the high-dimensional space, and obtain a reduced-dimensional data set; A model optimization unit is used to input the reduced-dimensional data set into a pre-trained random forest algorithm, output the recycling value and treatment priority of the waste, and evaluate the accuracy of the waste identification model in predicting the recycling value and treatment priority of the waste based on the historical data of waste classification and recycling through a feedback loop mechanism; if the performance of the waste identification model in predicting the recycling value and treatment priority does not meet the predetermined standards, optimize the parameters of the waste identification model through a genetic algorithm, and then apply the optimized parameters to the waste identification model; A model integration unit is used to integrate the optimized waste identification model into a system of a display interface, wherein the display interface is used to monitor and adjust the waste classification and recycling process in real time.
Citation Information
Patent Citations
Recycling bottle classification method based on deep learning model
CN112016462A
Garbage classification method and garbage classifier based on multi-label image recognition
CN113052247A
Garbage treatment method and device based on image recognition and electronic equipment
CN113610100A
Solid waste identification method based on spectral image
CN115600087A
Waste material data cleaning method and device, electronic equipment and storage medium
CN116303406A
