Garbage classification processing identification method based on AI algorithm analysis
By combining AI algorithms with multimodal sensing modules and multi-level waste classification and identification channels, the problem of inaccurate waste classification has been solved, and comprehensive and accurate waste classification and treatment have been achieved.
Patent Information
- Application Number
- CN202510855818.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-25
- Publication Date
- 2025-11-11
AI Technical Summary
Existing waste sorting methods rely on manual identification, which is inefficient and easily affected by subjective factors. Furthermore, automated systems ignore the harmfulness of waste and the urgency of its disposal, resulting in inaccurate and incomplete sorting.
Employing an AI-based multimodal sensing module and a multi-level waste classification and identification channel, combined with image sensors, gas sensors, weight sensors, and density sensors, multimodal perception is achieved to construct a waste sensing block. Through waste classification gating channels and multi-level waste classification and identification channels, the waste category, hazard level, and urgency of treatment are identified, generating a waste classification modality.
It has achieved comprehensive waste sorting, improved the accuracy and sustainability of waste sorting, and ensured that waste is precisely processed according to category, hazard level, and urgency.
Smart Images
Figure CN120929976A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing technology, and specifically to a waste classification and identification method based on AI algorithm analysis. Background Technology
[0002] With the acceleration of urbanization and the improvement of residents' living standards, the amount of waste generated has increased dramatically, making waste sorting and disposal a crucial issue for urban management and environmental protection. Traditional waste sorting methods mainly rely on manual identification and classification, which is not only inefficient but also easily influenced by subjective factors, leading to inaccurate sorting. Furthermore, while some existing automated waste sorting systems can identify waste categories, they often overlook the hazardousness of the waste and the urgency of its disposal, which to some extent limits the effectiveness and sustainability of waste sorting and disposal. Summary of the Invention
[0003] This application provides a waste sorting and identification method based on AI algorithm analysis, which solves the technical problem that the existing waste sorting and identification is not comprehensive enough and is limited to waste category identification.
[0004] This application provides a waste sorting and identification method based on AI algorithm analysis, the method comprising: The multimodal sensing module performs multimodal sensing on N unsorted wastes within the waste sorting area, constructing N waste sensing blocks. Based on these N waste sensing blocks, a processing order constraint is applied to the N unsorted wastes, resulting in the first unsorted waste, the second unprocessed waste, ..., the Nth unprocessed waste. The AI algorithm analysis module includes a waste sorting gating channel and a multi-level waste sorting identification channel, wherein the multi-level waste sorting identification channel includes a waste category identification unit, a waste hazard identification unit, and a waste processing urgency identification unit. Based on the N waste sensing blocks, the first waste sensing block corresponding to the first unsorted waste is extracted. Combining the waste sorting gating channel and the multi-level waste sorting identification channel, the first unsorted waste is sorted and identified to obtain a first waste sorting modality. Based on the waste sorting execution module, the first unsorted waste is sorted according to the first waste sorting modality, and the second unprocessed waste... the Nth unprocessed waste is further sorted according to the waste sorting gating channel, the multi-level waste sorting identification channel, and the waste sorting execution module.
[0005] Furthermore, based on the first waste sensing block, a first waste sorting gating coefficient is obtained according to the waste sorting gating channel; based on the first waste sorting gating coefficient and the first waste sensing block, a first waste category label is obtained according to the waste category identification unit; based on the first waste sorting gating coefficient and the first waste sensing block, a first waste hazard level is obtained according to the waste hazard identification unit; based on the first waste sorting gating coefficient and the first waste sensing block, a first waste disposal urgency level is obtained according to the waste disposal urgency identification unit; the first waste category label, the first waste hazard level, and the first waste disposal urgency level are fused to generate the first waste sorting modality.
[0006] Furthermore, the complexity of the first waste to be classified is evaluated based on the first waste sensing block to obtain a first waste complexity; the waste classification gating channel includes a waste classification gating network; the first waste complexity is input into the waste classification gating network to generate the first waste classification gating coefficient.
[0007] Furthermore, the waste category identification unit includes K waste category identification models, where K is a positive integer greater than 1; based on the first waste classification gating coefficient, feature activation is performed on the K waste category identification models to obtain multiple activated waste category identification models; the first waste perception block is input into the multiple activated waste category identification models to obtain multiple waste category identification results; confidence is evaluated based on the multiple waste category identification results to obtain multiple category confidence coefficients; based on the multiple category confidence coefficients, the multiple waste category identification results are filtered to maximize confidence, generating the first waste category label.
[0008] Furthermore, the waste hazard identification unit includes K waste hazard identification models; based on the first waste classification gating coefficient, feature activation is performed on the K waste hazard identification models to obtain multiple activated waste hazard identification models; the first waste sensing block is input into the multiple activated waste hazard identification models to obtain multiple waste hazard identification coefficients; the average value of the multiple waste hazard identification coefficients is calculated to obtain the first waste hazard level.
[0009] Furthermore, based on the N garbage odor information within the N garbage sensing blocks, odor anomaly detection is performed to obtain N odor anomaly feature information; based on the N odor anomaly feature information, anomaly depth evaluation is performed to obtain N garbage odor anomaly depth coefficients; based on the N garbage odor anomaly depth coefficients, the N garbage to be classified are sorted to obtain the first garbage to be classified, the second garbage to be processed, ... the Nth garbage to be processed.
[0010] Furthermore, the multimodal sensing module includes an image sensor, a gas sensor, a weight sensor, and a density sensor. Based on the multimodal sensing module, data is collected from the N types of waste to be classified, obtaining N waste image information, N waste odor information, N waste weight information, and N waste density information. The N waste image information is enhanced using an adaptive noise reduction function to obtain N enhanced waste images. Data quality enhancement is performed on the N waste odor information, the N waste weight information, and the N waste density information. Combined with the N enhanced waste images, the N waste sensing blocks are generated.
[0011] Furthermore, the adaptive noise reduction function is: ; Wherein, NRI(x,y) represents the garbage-enhanced image, LGC represents the brightness enhancement coefficient, L(x,y) represents the image brightness layer of the garbage image information, DGC represents the detail enhancement coefficient, and D(x,y) represents the image detail layer of the garbage image information.
[0012] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, the multimodal sensing module performs multimodal sensing on N pieces of waste to be classified within the waste sorting area, constructing N waste sensing blocks. Next, based on the N waste sensing blocks, a processing order constraint is applied to the N pieces of waste to be classified, resulting in the first piece of waste to be classified, the second piece of waste to be processed, ..., the Nth piece of waste to be processed. The AI algorithm analysis module includes a waste sorting gating channel and a multi-level waste sorting recognition channel. The multi-level waste sorting recognition channel includes a waste category recognition unit, a waste hazard recognition unit, and a waste processing urgency recognition unit. Next, based on the N waste sensing blocks, the first waste sensing block corresponding to the first piece of waste to be classified is extracted. Combining the waste sorting gating channel and the multi-level waste sorting recognition channel, the first piece of waste to be classified is then classified and identified, obtaining the first waste sorting modality. Finally, based on the waste sorting execution module, the first piece of waste to be classified is processed according to the first waste sorting modality, and the second piece of waste to be processed..., the Nth piece of waste to be processed is further processed according to the waste sorting gating channel, the multi-level waste sorting recognition channel, and the waste sorting execution module. This technology solves the problem that existing waste sorting and processing identification is not comprehensive enough and is limited to waste category identification, thus achieving the technical effect of improving the comprehensiveness of waste sorting. Attached Figure Description
[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0014] Figure 1 A schematic diagram of a waste sorting and identification method based on AI algorithm analysis is provided in an embodiment of this application; Figure 2 This is a schematic diagram illustrating the process of obtaining the first waste classification modality in a waste classification and identification method based on AI algorithm analysis provided in an embodiment of this application. Detailed Implementation
[0015] This application provides a waste sorting and identification method based on AI algorithm analysis, which solves the technical problem that the existing waste sorting and identification is not comprehensive enough and is limited to waste category identification.
[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.
[0017] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.
[0018] Examples, such as Figure 1 As shown in the figure, this application provides a waste sorting and identification method based on AI algorithm analysis. The method is applied to a waste sorting robot, which includes a multimodal sensing module, an AI algorithm analysis module, and a waste sorting execution module. The method includes: The multimodal sensing module performs multimodal sensing on N types of waste to be classified within the waste sorting area, thereby constructing N waste sensing blocks.
[0019] The multimodal sensing module integrates various sensors (such as image sensors, gas sensors, weight sensors, and density sensors). These sensors work together to capture multiple features of waste, including its shape, color, weight, and material. By using the multimodal sensing module to comprehensively and meticulously perceive N pieces of waste to be sorted within a waste sorting area, N independent waste sensing blocks are constructed. Each block contains multidimensional information about the corresponding waste, such as its shape, color, material, and weight.
[0020] Furthermore, based on the multimodal sensing module's multimodal sensing of N types of waste to be classified within the waste sorting area, N waste sensing blocks are constructed, including: The multimodal sensing module includes an image sensor, a gas sensor, a weight sensor, and a density sensor. Based on the multimodal sensing module, data is collected from the N types of waste to be classified, obtaining N waste image information, N waste odor information, N waste weight information, and N waste density information. The N waste image information is enhanced using an adaptive noise reduction function to obtain N enhanced waste images. Data quality enhancement is performed on the N waste odor information, the N waste weight information, and the N waste density information. Combined with the N enhanced waste images, the N waste sensing blocks are generated.
[0021] The multimodal sensing module integrates an image sensor, a gas sensor, a weight sensor, and a density sensor. Using this module, data is collected from N pieces of waste to be sorted, yielding N images, N odor data, N weight data, and N density data for each piece of waste. Specifically, the image sensor captures the appearance features of each piece of waste, generating N images; the gas sensor collects the odor emitted by the waste, generating N odor data; and the weight and density sensors measure the weight and density of each piece of waste, respectively, generating N weight and N density data.
[0022] Raw data often contains noise interference. To improve the accuracy and reliability of the data, an adaptive denoising function can be used to enhance N garbage image information, resulting in N enhanced garbage images. Simultaneously, data quality enhancement processing is performed on N garbage odor information, N garbage weight information, and N garbage density information, including data cleaning, missing value imputation, and outlier detection and handling, ensuring that all input data is high-quality and reliable. Combining the N enhanced garbage images with other enhanced sensory information, N garbage sensing blocks are generated. Each garbage sensing block is an independent data unit containing multi-dimensional feature information, providing comprehensive and accurate data support for subsequent classification, identification, and processing.
[0023] Furthermore, the adaptive noise reduction function is: ; Wherein, NRI(x,y) represents the garbage-enhanced image, LGC represents the brightness enhancement coefficient, L(x,y) represents the image brightness layer of the garbage image information, DGC represents the detail enhancement coefficient, and D(x,y) represents the image detail layer of the garbage image information.
[0024] The adaptive denoising function effectively enhances junk image information by comprehensively considering information from both the image brightness layer and the image detail layer. Specifically, the image brightness layer L(x, y) reflects the overall brightness distribution of the junk image, while the image detail layer D(x, y) contains detailed information such as edges and textures in the image. LGC is used to adjust the weights of the image brightness layer L(x, y), while DGC is used to adjust the weights of the image detail layer D(x, y). The brightness enhancement coefficient and detail enhancement coefficient are combined with the corresponding image layers, and the junk-enhanced image NRI(x, y) is obtained through a weighted summation.
[0025] Based on the N waste sensing blocks, the processing order of the N wastes to be classified is constrained to obtain the first waste to be classified, the second waste to be processed, ... the Nth waste to be processed.
[0026] Processing order constraints refer to assigning a processing order to each piece of waste to be sorted based on its different characteristics (such as type, hazard, priority, etc.). By comprehensively evaluating the information in N waste sensing blocks, the key characteristics of each piece of waste can be identified, and a reasonable processing order can be generated based on these characteristics. According to this order, the waste is sequentially marked as the first piece of waste to be sorted, the second piece of waste to be processed, and so on, up to the Nth piece of waste to be processed.
[0027] Furthermore, based on the N waste sensing blocks, the processing order of the N wastes to be classified is constrained to obtain the first waste to be classified, the second waste to be processed, ..., the Nth waste to be processed, including: Odor anomaly detection is performed based on N garbage odor information within the N garbage sensing blocks to obtain N odor anomaly feature information; anomaly depth evaluation is performed based on the N odor anomaly feature information to obtain N garbage odor anomaly depth coefficients; the N garbage to be classified are sorted based on the N garbage odor anomaly depth coefficients to obtain the first garbage to be classified, the second garbage to be processed, ... the Nth garbage to be processed.
[0028] Odor anomaly detection is used to determine whether the odor of waste is abnormal. By comparing the odor information of each piece of waste with a preset odor database, it identifies whether there is an abnormal odor, such as irritating, corrosive, toxic, or harmful. By performing odor anomaly detection on N pieces of waste odor information, N odor anomaly feature information can be generated. The odor anomaly feature information indicates whether the odor of each piece of waste is abnormal, and the degree of abnormality.
[0029] Anomaly depth evaluation is performed based on N odor anomaly characteristics. Optionally, the odor concentration data of the waste is acquired and checked to see if it exceeds a preset abnormal concentration threshold. If it exceeds, it is marked as abnormal; otherwise, it is marked as normal. Anomaly depth values are mapped to the odor concentration and type. For example, if the odor concentration is greater than the threshold, a score can be set according to the concentration range: concentration > 80%, anomaly depth coefficient of 1.0; 60% < concentration ≤ 80%, anomaly depth coefficient of 0.7; 40% < concentration ≤ 60%, anomaly depth coefficient of 0.5; concentration ≤ 40%, anomaly depth coefficient of 0.2. Furthermore, different weighting coefficients are applied to different gas types (such as harmful gases and putrid gases). For example, the weighting coefficient for hydrogen sulfide is 2.0, while for ordinary rotten odors it is 1.0. For each piece of waste odor information, a corresponding odor anomaly depth coefficient is calculated, resulting in N waste odor anomaly depth coefficients. These coefficients reflect the severity of the waste odor anomaly.
[0030] The waste to be sorted is sorted according to N odor abnormality depth coefficients. Waste with higher odor abnormality depth coefficients is processed first, and so on, until the first waste to be sorted, the second waste to be processed, and so on, until the Nth waste to be processed.
[0031] The AI algorithm analysis module includes a waste sorting gating channel and a multi-level waste sorting identification channel. The multi-level waste sorting identification channel includes a waste category identification unit, a waste hazard identification unit, and a waste disposal urgency identification unit.
[0032] The AI algorithm analysis module includes a waste sorting gating channel and a multi-level waste sorting recognition channel. The waste sorting gating channel is responsible for the initial screening and control of waste, guiding it into the appropriate sorting path based on input waste information (such as images, odors, and weight). The multi-level waste sorting recognition channel includes a waste category recognition unit, a waste hazard identification unit, and a waste disposal urgency identification unit. The waste category recognition unit identifies basic waste categories, such as recyclable waste, non-recyclable waste, and hazardous waste. The waste hazard identification unit assesses the degree of hazard of waste, especially identifying waste that may release toxic substances or harm the environment. The waste disposal urgency identification unit assesses the priority of waste disposal. Through the coordinated operation of the waste sorting gating channel and the multi-level waste sorting recognition channel, the AI algorithm analysis module can perform refined waste processing, identifying the type, hazard level, and urgency of waste disposal, thereby achieving efficient and accurate waste sorting and processing.
[0033] Based on the N waste sensing blocks, the first waste sensing block corresponding to the first waste to be classified is extracted. The first waste to be classified is then classified and identified by combining the waste classification gating channel and the multi-level waste classification recognition channel to obtain the first waste classification mode.
[0034] Extract the first waste sensing block corresponding to the first waste to be classified from N waste sensing blocks. The first waste sensing block contains all the multidimensional information of the first waste, such as the image features, odor intensity, weight, density, etc.
[0035] The waste sorting gating channel plays a controlling and guiding role, responsible for directing the extracted waste information (i.e., the first waste sensing block) to the appropriate sorting path. Specifically, it first determines whether the waste meets certain basic conditions or preset rules, deciding whether to pass it to the multi-level waste sorting identification channel for further analysis. The multi-level waste sorting identification channel will analyze the type, hazard level, and urgency of the waste based on the information in the first waste sensing block. After the above steps, a comprehensive classification information for the first waste to be sorted, i.e., the first waste sorting modality, is generated. The first waste sorting modality contains key information such as the specific category of waste, hazard level, and urgency of treatment, providing an important basis for subsequent processing and decision-making.
[0036] Furthermore, such as Figure 2 As shown, based on the N waste sensing blocks, the first waste sensing block corresponding to the first waste to be classified is extracted. Combined with the waste classification gating channel and the multi-level waste classification recognition channel, the first waste to be classified is classified and identified to obtain the first waste classification modality, including: Based on the first waste sensing block, a first waste sorting gating coefficient is obtained according to the waste sorting gating channel; based on the first waste sorting gating coefficient and the first waste sensing block, a first waste category label is obtained according to the waste category identification unit; based on the first waste sorting gating coefficient and the first waste sensing block, a first waste hazard level is obtained according to the waste hazard identification unit; based on the first waste sorting gating coefficient and the first waste sensing block, a first waste disposal urgency level is obtained according to the waste disposal urgency identification unit; the first waste sorting modality is generated by fusing the first waste category label, the first waste hazard level, and the first waste disposal urgency level.
[0037] The first waste sensing block is input into the waste sorting gating channel. After analysis, the waste sorting gating channel outputs the first waste sorting gating coefficient, which reflects the complexity of the waste in the sorting process. Combining the first waste sorting gating coefficient with image and shape information in the first waste sensing block, the waste category recognition unit identifies the waste and obtains the first waste category label, which represents the specific category of the first type of waste. Using the waste hazard recognition unit, the hazard level of the waste is assessed based on the first waste sorting gating coefficient and information such as composition and odor in the first waste sensing block, resulting in the first waste hazard level, which represents the degree of hazard of the waste. Using the waste disposal urgency recognition unit, the urgency of waste disposal is assessed based on the first waste sorting gating coefficient and information such as storage time and quantity in the first waste sensing block, resulting in the first waste disposal urgency, which represents the priority of waste disposal. The first waste category label, the first waste hazard level, and the first waste disposal urgency are fused to generate a first waste sorting modality containing this key information.
[0038] Furthermore, based on the aforementioned waste sorting gating channel, a first waste sorting gating coefficient is obtained, including: The complexity of the first waste to be classified is evaluated based on the first waste sensing block to obtain a first waste complexity; the waste classification gating channel includes a waste classification gating network; the first waste complexity is input into the waste classification gating network to generate the first waste classification gating coefficient.
[0039] The complexity of the first waste sensing block is evaluated to obtain the first waste complexity. Specifically, in image feature analysis, the complexity is determined by calculating the shape complexity (such as irregular shapes or sharp edges) and texture complexity (such as complex textures or uneven surfaces) of the waste. In terms of odor features, the odor concentration and types of waste are measured by gas sensors. Higher odor concentration or multiple odors indicate higher waste processing complexity. Weight and density analysis are also included in the evaluation, and the difficulty of sorting and processing waste is judged by its greater weight and higher density. Taking all these factors into account, different weights are assigned to each waste feature, and a comprehensive complexity score is calculated as the first waste complexity.
[0040] The waste sorting gating system includes a waste sorting gating network trained using a neural network. The core task of this network is to generate a first waste sorting gating coefficient based on the input complexity of the first type of waste. Specifically, the network learns the complexity characteristics of waste through training and automatically adjusts weights and parameters using a large amount of training data to optimize the waste sorting control strategy. After inputting the first waste complexity into the gating network, it outputs the first waste sorting gating coefficient based on preset sorting rules and trained weights. This coefficient represents the number of activated models, dynamically determined by the perceived characteristics of the waste (such as image complexity, odor intensity, weight, etc.).
[0041] Furthermore, based on the first waste sorting gating coefficient and the first waste sensing block, a first waste category label is obtained according to the waste category identification unit, including: The waste category identification unit includes K waste category identification models, where K is a positive integer greater than 1; based on the first waste classification gating coefficient, feature activation is performed on the K waste category identification models to obtain multiple activated waste category identification models; the first waste perception block is input into the multiple activated waste category identification models to obtain multiple waste category identification results; confidence is evaluated based on the multiple waste category identification results to obtain multiple category confidence coefficients; based on the multiple category confidence coefficients, the multiple waste category identification results are filtered to maximize confidence, generating the first waste category label.
[0042] The waste category identification unit consists of K waste category identification models, where K is a positive integer greater than 1. Based on a first waste classification gating coefficient, feature activation is performed on the K waste category identification models. The first waste classification gating coefficient determines which identification models should be activated for classification. More complex waste will activate more identification models, while simple waste may only activate a few models. For example, if the first waste classification gating coefficient is 4, then 4 of the K waste category identification models will be activated, resulting in 4 activated waste category identification models. The first waste perception block is input into the multiple activated waste category identification models for classification. Each activated waste category identification model will output a value based on the multi-dimensional perception features of the waste (such as image, smell, weight, density, etc.). Waste category identification results; the confidence level of multiple waste category identification results is evaluated. The confidence level of each identification result can be calculated based on the output value and accuracy of each model. Generally, if the output of a certain identification model is the probability value of a certain category (e.g., the probability of the recyclable category is 0.85), then the probability value itself can be used as the confidence level of that category, i.e., the category confidence coefficient. Based on multiple category confidence coefficients, the confidence level of multiple waste category identification results is maximized, that is, the identification result with the highest confidence level is selected as the final classification result, and the first waste category label is generated, that is, an accurate classification label is assigned to the waste, such as recyclable, hazardous waste, or non-recyclable, thereby ensuring the accuracy and reliability of waste classification.
[0043] Furthermore, based on the first waste sorting threshold coefficient and the first waste sensing block, the first waste hazard level is obtained according to the waste hazard identification unit, including: The waste hazard identification unit includes K waste hazard identification models; based on the first waste classification gating coefficient, the K waste hazard identification models are feature activated to obtain multiple activated waste hazard identification models; the first waste sensing block is input into the multiple activated waste hazard identification models to obtain multiple waste hazard identification coefficients; the average of the multiple waste hazard identification coefficients is calculated to obtain the first waste hazard level.
[0044] The waste hazard identification unit consists of K waste hazard identification models, where K is a positive integer greater than 1. Each model is responsible for assessing the hazard of waste and identifying whether it contains hazardous substances such as toxic gases, putrefactive substances, and chemicals. The waste hazard identification models can be built based on machine learning algorithms and, after training, can assess the potential hazards of waste. Based on the first waste classification gating coefficient, feature activation is performed on the K waste hazard identification models to obtain multiple activated waste hazard identification models. The first waste sensing block (containing sensing data such as image, odor, weight, and density) is input into the activated waste hazard identification models for hazard analysis. Each activated model calculates the waste hazard identification coefficient based on the characteristics of the sensing block. The waste hazard identification coefficient reflects the potential hazard level of the waste. After obtaining multiple waste hazard identification coefficients, the average of these coefficients is calculated to obtain the first waste hazard level, which reflects the overall hazard level of the waste.
[0045] Based on the first waste classification gating coefficient and the first waste sensing block, the process of obtaining the first waste disposal urgency through the waste disposal urgency identification unit is similar to the process of obtaining the waste hazard level through the waste hazard identification unit. Specifically, the waste disposal urgency identification unit includes K waste disposal urgency identification models, where K is a positive integer greater than 1. Each model is responsible for assessing the urgency of waste disposal, i.e., the waste disposal priority. These models determine whether waste needs urgent disposal based on multi-dimensional sensing data of waste (such as images, odors, weight, density, etc.). For example, some hazardous waste may need to be disposed of immediately, while ordinary waste can be disposed of later. Based on the first waste classification gating coefficient, feature activation is performed on the K waste disposal urgency identification models. The first waste sensing block is input into the activated waste disposal urgency identification models for urgency assessment. Each activated model calculates a waste disposal urgency identification coefficient based on the characteristics of the waste. This coefficient represents the waste disposal priority or urgency; the higher the value, the higher the waste disposal priority. After obtaining multiple waste disposal urgency identification coefficients, the average of the multiple waste disposal urgency identification coefficients is calculated to obtain the first waste disposal urgency.
[0046] Based on the waste sorting execution module, the first waste to be sorted is sorted according to the first waste sorting mode, and the second waste to be processed...the Nth waste to be processed is further sorted according to the waste sorting gate channel, the multi-level waste sorting identification channel and the waste sorting execution module.
[0047] Based on the waste sorting execution module, the first waste to be sorted is processed through a first waste sorting mode. This first mode generates a classification result by analyzing the waste's sensory information (such as images, odors, weight, etc.) through a waste sorting gating channel and a multi-level waste sorting recognition channel. This result includes category labels, hazard level, and processing urgency information, guiding the waste sorting and processing. The waste sorting execution module then processes the waste based on this information, such as classifying it as recyclable, hazardous, or non-recyclable, and performs corresponding operations. After processing the first batch of waste, the waste sorting gating channel and multi-level waste sorting recognition channel are used to process subsequent waste, up to the Nth batch, ensuring that each piece of waste receives appropriate processing based on its complexity, hazard level, and urgency.
[0048] In summary, the embodiments of this application have at least the following technical effects: First, the multimodal sensing module performs multimodal sensing on N pieces of waste to be classified within the waste sorting area, constructing N waste sensing blocks. Next, based on the N waste sensing blocks, a processing order constraint is applied to the N pieces of waste to be classified, resulting in the first piece of waste to be classified, the second piece of waste to be processed, ..., the Nth piece of waste to be processed. The AI algorithm analysis module includes a waste sorting gating channel and a multi-level waste sorting recognition channel. The multi-level waste sorting recognition channel includes a waste category recognition unit, a waste hazard recognition unit, and a waste processing urgency recognition unit. Next, based on the N waste sensing blocks, the first waste sensing block corresponding to the first piece of waste to be classified is extracted. Combining the waste sorting gating channel and the multi-level waste sorting recognition channel, the first piece of waste to be classified is then classified and identified, obtaining the first waste sorting modality. Finally, based on the waste sorting execution module, the first piece of waste to be classified is processed according to the first waste sorting modality, and the second piece of waste to be processed..., the Nth piece of waste to be processed is further processed according to the waste sorting gating channel, the multi-level waste sorting recognition channel, and the waste sorting execution module. This technology solves the problem that existing waste sorting and processing identification is not comprehensive enough and is limited to waste category identification, thus achieving the technical effect of improving the comprehensiveness of waste sorting.
[0049] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.
[0050] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
[0051] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.
Claims
1. A waste sorting and identification method based on AI algorithm analysis, characterized in that, The method is applied to a waste sorting and processing robot, which includes a multimodal sensing module, an AI algorithm analysis module, and a waste sorting execution module. The method includes: The multimodal sensing module performs multimodal sensing on N types of waste to be classified within the waste sorting area, and constructs N waste sensing blocks. Based on the N waste sensing blocks, the processing order of the N wastes to be classified is constrained to obtain the first waste to be classified, the second waste to be processed, ... the Nth waste to be processed; The AI algorithm analysis module includes a waste sorting gating channel and a multi-level waste sorting identification channel, wherein the multi-level waste sorting identification channel includes a waste category identification unit, a waste hazard identification unit, and a waste disposal urgency identification unit; Based on the N waste sensing blocks, the first waste sensing block corresponding to the first waste to be classified is extracted. The first waste to be classified is then classified and identified by combining the waste classification gating channel and the multi-level waste classification recognition channel to obtain the first waste classification mode. Based on the waste sorting execution module, the first waste to be sorted is sorted according to the first waste sorting mode, and the second waste to be processed...the Nth waste to be processed is further sorted according to the waste sorting gate channel, the multi-level waste sorting identification channel and the waste sorting execution module.
2. The waste sorting and identification method based on AI algorithm analysis as described in claim 1, characterized in that, Based on the N waste sensing blocks, the first waste sensing block corresponding to the first waste to be classified is extracted. Then, the first waste to be classified is classified and identified using the waste classification gating channel and the multi-level waste classification recognition channel to obtain a first waste classification modality, including: Based on the first waste sensing block, a first waste sorting gating coefficient is obtained according to the waste sorting gating channel; Based on the first waste classification gating coefficient and the first waste sensing block, a first waste category label is obtained according to the waste category identification unit; Based on the first waste classification gating coefficient and the first waste sensing block, the first waste hazard level is obtained according to the waste hazard identification unit; Based on the first waste sorting gating coefficient and the first waste sensing block, the first waste disposal urgency is obtained according to the waste disposal urgency identification unit; The first waste classification modality is generated by integrating the first waste category label, the first waste hazard level, and the first waste disposal urgency.
3. The waste sorting and identification method based on AI algorithm analysis as described in claim 2, characterized in that, Based on the first waste sensing block, and according to the waste sorting gating channel, a first waste sorting gating coefficient is obtained, including: The complexity of the first waste to be classified is evaluated based on the first waste sensing block to obtain the first waste complexity. The waste sorting gate control channel includes a waste sorting gate control network; The first waste complexity is input into the waste sorting gating network to generate the first waste sorting gating coefficient.
4. The waste sorting and identification method based on AI algorithm analysis as described in claim 2, characterized in that, Based on the first waste sorting threshold coefficient and the first waste sensing block, a first waste category label is obtained according to the waste category identification unit, including: The waste category identification unit includes K waste category identification models, where K is a positive integer greater than 1; Based on the first waste classification gating coefficient, feature activation is performed on the K waste category recognition models to obtain multiple activated waste category recognition models; The first waste sensing block is input into the multiple activated waste category recognition models to obtain multiple waste category recognition results; Based on the identification results of the multiple waste categories, a confidence evaluation is performed to obtain confidence coefficients for multiple categories; Based on the confidence coefficients of the multiple categories, the identification results of the multiple waste categories are filtered to maximize confidence, and the first waste category label is generated.
5. The waste sorting and identification method based on AI algorithm analysis as described in claim 2, characterized in that, Based on the first waste sorting threshold coefficient and the first waste sensing block, the first waste hazard level is obtained according to the waste hazard identification unit, including: The waste hazard identification unit includes K waste hazard identification models; Based on the first waste classification gating coefficient, feature activation is performed on the K waste hazard identification models to obtain multiple activated waste hazard identification models; The first waste sensing block is input into the multiple activated waste hazard identification models to obtain multiple waste hazard identification coefficients; The first waste hazard level is obtained by averaging the multiple waste hazard identification coefficients.
6. The waste sorting and identification method based on AI algorithm analysis as described in claim 1, characterized in that, Based on the N waste sensing blocks, the processing order of the N wastes to be classified is constrained to obtain the first waste to be classified, the second waste to be processed, ..., the Nth waste to be processed, including: Odor anomaly detection is performed based on N garbage odor information within the N garbage sensing blocks to obtain N odor anomaly feature information; Based on the N odor anomaly feature information, anomaly depth evaluation is performed to obtain N garbage odor anomaly depth coefficients; The N wastes to be classified are sorted according to the N waste odor anomaly depth coefficients to obtain the first waste to be classified, the second waste to be processed, ... the Nth waste to be processed.
7. The waste sorting and identification method based on AI algorithm analysis as described in claim 1, characterized in that, The multimodal sensing module performs multimodal sensing on N types of waste to be classified within the waste sorting area, constructing N waste sensing blocks, including: The multimodal sensing module includes an image sensor, a gas sensor, a weight sensor, and a density sensor; According to the multimodal sensing module, data is collected from the N types of waste to be classified to obtain N waste image information, N waste odor information, N waste weight information and N waste density information; The N garbage image information are enhanced according to the adaptive noise reduction function to obtain N garbage enhanced images; Data quality enhancement is performed on the N garbage odor information, the N garbage weight information, and the N garbage density information. Combined with the N garbage enhanced images, the N garbage sensing blocks are generated.
8. The waste sorting and identification method based on AI algorithm analysis as described in claim 7, characterized in that, The adaptive noise reduction function is: ； Wherein, NRI(x,y) represents the garbage-enhanced image, LGC represents the brightness enhancement coefficient, L(x,y) represents the image brightness layer of the garbage image information, DGC represents the detail enhancement coefficient, and D(x,y) represents the image detail layer of the garbage image information.