A Neuro-Fuzzy Based Adaptive Garbage Cleaning Method and System

Abstract

The present invention relates to the technical field of garbage cleaning, and in particular to an adaptive garbage cleaning method and system based on neuro-fuzzy. The method includes constructing a multi-modal garbage image data set and dividing the data set; fusing fuzzy clustering and deep learning features to train an object detection model; performing fuzzy processing on images to determine garbage categories; establishing a fuzzy rule base to achieve real-time decision-making; using a hybrid optimization strategy to train the model; estimating the volume and mass of garbage based on three-dimensional perception technology; using a fuzzy PID controller to adjust the power of the cleaning device; deploying an edge-cloud collaborative architecture to update the rule base; the system includes a multi-modal data acquisition unit, a hybrid computing unit, a dynamic power control unit, and a cloud management platform. The invention can accurately identify garbage categories, optimize the power of the cleaning device, achieve the coordination of energy consumption and efficiency, improve the adaptability and real-time decision-making of the system, and effectively solve the deficiencies of traditional garbage cleaning technologies.

Description

Technical Field

[0001] The present invention relates to the technical field of garbage cleaning, and specifically to an adaptive garbage cleaning method and system based on neuro-fuzzy. Background Art

[0002] In today's society, garbage cleaning plays a crucial role in urban environmental maintenance, resource recycling, and public health protection. With the continuous expansion of urban scale and the continuous improvement of residents' living standards, the output and types of garbage are increasing rapidly, which poses higher requirements for garbage cleaning technology. However, the existing garbage cleaning technology has the following problems to be solved urgently:

[0003] Problems in garbage recognition and classification: Traditional garbage cleaning equipment mainly relies on manual recognition or simple image recognition technology to distinguish garbage types; manual recognition is inefficient and easily affected by subjective factors and fatigue, and it is difficult to ensure accuracy and consistency in large-scale garbage cleaning scenarios; while simple image recognition technology faces many challenges when dealing with garbage images in complex backgrounds; in poor light conditions (such as at night, on cloudy days, or in shadow areas), it is difficult to extract image features and misjudgments are likely to occur; when there is partial occlusion, overlap, or similar morphology of garbage, existing algorithms are also difficult to accurately distinguish different types of garbage, resulting in incorrect classification and treatment of garbage, affecting subsequent recycling or environmental protection disposal;

[0004] Imbalance between cleaning efficiency and energy consumption: Current cleaning devices lack intelligence in power regulation and usually operate at a fixed power or a limited number of power levels; they cannot achieve precise power matching when facing different quantities and types of garbage; when encountering a small amount of light garbage, high-power operation will cause great waste of energy; while when dealing with a large amount of heavy garbage or tightly packed garbage, insufficient power will greatly reduce the cleaning effect, prolong the cleaning time, and reduce the overall cleaning efficiency; this imbalance between efficiency and energy consumption not only increases the operating cost but also does not conform to the concept of sustainable development;

[0005] Lack of real-time adaptive decision-making ability: The garbage cleaning environment is complex and changeable, and the garbage distribution and types vary significantly in different areas (such as commercial areas, residential areas, parks, etc.) and may change at any time; existing cleaning systems often cannot perceive these changes in real time and make corresponding decision adjustments; for example, when a large amount of garbage suddenly accumulates in a certain area, traditional systems cannot quickly optimize the cleaning path and adjust the working mode of the cleaning device, resulting in a lag in the cleaning work and affecting the environmental quality; in addition, in the face of newly emerging garbage types or special cleaning scenarios, the system lacks self-learning and adaptation abilities and is difficult to meet the ever-changing actual needs;

[0006] Insufficient data processing and utilization: With the development of sensor technology, the amount of data that can be collected during the garbage cleaning process has increased significantly, including garbage images, location information, operation status data of cleaning equipment, etc.; however, the existing cleaning systems have serious deficiencies in data processing and cannot fully exploit the value of this data; a large amount of data is simply stored or only used for basic status monitoring and fails to be effectively used to optimize cleaning strategies, improve equipment performance, and refine the garbage treatment process; at the same time, there are also obstacles in data transmission and sharing, and it is difficult to synergistically utilize the data between different devices and systems, forming information silos, which limits the improvement of overall cleaning efficiency.

[0007] Therefore, in view of the above problems, an adaptive garbage cleaning method and system based on neuro-fuzzy is proposed. Summary of the Invention

[0008] The purpose of the present invention is to provide an adaptive garbage cleaning method and system based on neuro-fuzzy to solve the problems raised in the above background technology.

[0009] To achieve the above purpose, the present invention provides the following technical solutions:

[0010] An adaptive garbage cleaning method based on neuro-fuzzy includes the following steps:

[0011] S1. Construct a multi-modal garbage image dataset by annotating garbage types and dividing the training set, validation set, and test set;

[0012] S2. Train an object detection model based on multi-modal feature fusion, fuse fuzzy clustering features and deep learning features, and dynamically optimize the classification boundary;

[0013] S3. Perform fuzzification processing on the garbage image, generate a membership matrix by combining the fuzzy clustering algorithm, and set a threshold to determine the garbage category;

[0014] S4. Establish a fuzzy rule base and achieve real-time decision-making through a fuzzy inference model;

[0015] S5. Train the model using a hybrid optimization strategy, combine global search and second-order accuracy tuning to improve the model efficiency and convergence speed;

[0016] S6. Based on three-dimensional perception technology, estimate the volume and mass of garbage in real time, and match physical parameters by combining with a density database;

[0017] S7. Dynamically adjust the power of the cleaning device through a fuzzy PID controller to achieve coordinated optimization of energy consumption and efficiency;

[0018] S8. Deploy an edge-cloud collaborative architecture to support real-time decision-making and incremental update of the rule base.

[0019] As a preferred solution, in step S2, the multi-modal feature fusion specifically includes:

[0020] The membership matrix generated by FCM fuzzy clustering and the feature map output by YOLOv8 are cascaded at the channel level, where is the membership matrix generated by FCM fuzzy clustering, is the number of clusters, is the number of samples, is the feature map output by YOLOv8, is the height of the feature map, is the width of the feature map, is the number of channels of the feature map, and the feature weights are calculated through a dynamic weight allocation module. The formula is:

[0021] , where is the feature weight of the th cluster, is the membership of the th cluster, is the membership of the th cluster, is the temperature coefficient, is the number of clusters;

[0022] The weighted fusion feature is input into the detection head for classification, where is the weighted fusion feature, is the feature map corresponding to the th cluster.

[0023] As a preferred solution, the fuzzy clustering algorithm in step S3 is FCM, and its membership update formula is:

[0024] , where is the membership of the th sample belonging to the th cluster, is the th sample, is the th cluster center, represents the th cluster center, is the fuzzy index;

[0025] Set , and when the membership threshold , it is determined as "big garbage", and the classification accuracy is verified to be > 92% through the confusion matrix.

[0026] As a preferred solution, the hybrid optimization strategy in step S5 includes:

[0027] Grid search stage: Traverse combinations within the parameter space to screen out candidate parameters with an error lower than the threshold , where and are the parameters to be optimized, is the error threshold;

[0028] L-M optimization stage: Perform second-order optimization based on the candidate parameters, and the update formula is:

[0029] , where is the update amount of the parameter, is the Jacobian matrix of the error with respect to the parameter, is the Jacobian matrix of the transposed matrix, is the damping factor, is the identity matrix, is the error vector, and the initial value of the damping factor has a dynamic adjustment range of .

[0030] As a preferred solution, in step S6, the three-dimensional volume estimation formula is:

[0031] , where is the three-dimensional volume of the garbage; is the number of pixels; is the physical size of the pixel in the direction, , is the physical size of the pixel in the direction, , is the physical width of the sensor, is the width of the image, is the physical height of the sensor, is the height of the image; is the depth value of the th pixel; The quality is calculated as , is the mass of the garbage, is the density of the garbage, and the density is matched from a preset database, and the database contains density values of 20 types of garbage.

[0032] As a preferred solution, in step S7, the rule base of the fuzzy PID controller includes:

[0033] Rule 1: is "Dazheng" and is "Zhengkuai", THEN , where is the error at the current moment; is the rate of change of the error; is the adjustment amount of the power; is the proportionality coefficient; is the differential coefficient;

[0034] Rule 2: is "Zhong" and is "Zero", THEN ;

[0035] Rule 3: is "Xiaofu" and is "Fuman", , where is the integral coefficient; is the integral of the error;

[0036] The membership function adopts a Gaussian type, defined as , where is the error belongs to the membership degree of "Large", is the center of the Gaussian membership function, is the standard deviation of the Gaussian membership function, , , response time < 200ms.

[0037] As a preferred solution, the edge-cloud collaborative architecture in step S8 includes:

[0038] Deploy a lightweight TSK fuzzy inference model at the edge side, and its rule output is:

[0039] , where is the output of the th rule, is the input vector; is the constant term of the th rule; is the coefficient of the th rule for the th input variable, updated through incremental learning, is the number of input variables;

[0040] The cloud performs density database expansion every month, adding the density value of the garbage category , and synchronizes it to the edge device through OTA.

[0041] An adaptive garbage cleaning system based on neuro-fuzzy, comprising:

[0042] Multi-modal data acquisition unit: Integrating an RGB-D camera and a lidar to collect image and depth information;

[0043] Hybrid computing unit: Deploying an adaptive garbage cleaning method based on neuro-fuzzy, including an FPGA-accelerated fuzzy inference module. The FPGA-accelerated fuzzy inference module adopts a parallel architecture to perform hardware acceleration on the fuzzy inference process. The single-frame processing delay is <50ms and the power consumption is <5W;

[0044] Dynamic power control unit: Supporting power regulation from 50W to 150W, and the suction non-linear mapping formula is , where is the suction force, is the power, is the coefficient, is the exponent, , , and the maximum suction force ≥ 200N;

[0045] Cloud management platform: Providing rule library version control and incremental learning interfaces.

[0046] It can be seen from the technical solutions provided by the present invention above that an adaptive garbage cleaning method and system based on neuro-fuzzy provided by the present invention have the following beneficial effects:

[0047] Improving the accuracy of garbage recognition and classification: By constructing a multi-modal garbage image dataset, fusing fuzzy clustering features and deep learning features for target detection model training, and dynamically optimizing the classification boundary, different types of garbage can be recognized more accurately; Using the membership matrix generated by FCM fuzzy clustering and the feature map output by YOLOv8 for channel concatenation, and calculating the feature weights through a dynamic weight allocation module, the model can fully exploit the fuzzy information and deep learning features in the data to improve the classification accuracy; Determining the garbage category by setting the membership threshold of the fuzzy clustering algorithm, and verifying that the classification accuracy is greater than 92% through a confusion matrix, effectively improving the reliability of garbage recognition;

[0048] Optimize the power adjustment of the cleaning device and reduce energy consumption: Adopt a fuzzy PID controller to dynamically adjust the power of the cleaning device. According to the real-time situation of the garbage (such as volume, mass, etc.) and the working state of the cleaning device, adjust the power size in real time to achieve the collaborative optimization of energy consumption and efficiency; By establishing a fuzzy rule base, adjust the power according to factors such as error and error change rate, so that the cleaning device can ensure the cleaning effect and avoid unnecessary energy consumption when facing different garbage scenarios; The suction non-linear mapping formula realizes the reasonable matching of power and suction. In the power adjustment range of 50W - 150W, the maximum suction force ≥ 200N, ensuring efficient cleaning under different garbage loads;

[0049] Implement real-time decision-making and system adaptive update: Deploy an edge-cloud collaborative architecture. Deploy a lightweight TSK fuzzy inference model at the edge side to support real-time decision-making and be able to quickly respond to various situations during the garbage cleaning process; The cloud expands the density database monthly and synchronizes it to the edge device through OTA to achieve incremental update of the rule base, enabling the system to continuously adapt to newly emerging garbage categories and scenario changes; With the emergence of new garbage types, the cloud updates the density database and synchronizes it to the edge device to ensure the accuracy of garbage quality estimation, and then optimize the cleaning strategy to improve the overall adaptability and intelligence level of the system;

[0050] Improve the system computing efficiency and stability: The fuzzy inference module accelerated by FPGA in the hybrid computing unit adopts a parallel architecture to perform hardware acceleration on the fuzzy inference process. The single-frame processing delay < 50ms and the power consumption < 5W, greatly improving the system computing efficiency and ensuring the stability and fast response ability when processing a large amount of garbage image data in real time; This hardware acceleration method not only improves the speed of garbage recognition and decision-making, but also reduces the energy consumption of the system and enhances the feasibility and reliability of the system in practical applications. Brief Description of the Drawings

[0051] Figure 1 It is a step schematic diagram of a neural fuzzy-based adaptive garbage cleaning method of the present invention;

[0052] Figure 2 It is a structural schematic diagram of a neural fuzzy-based adaptive garbage cleaning system of the present invention. Detailed Embodiments

[0053] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0054] In order to better understand the above technical solutions, the above technical solutions will be described in detail below in conjunction with the specification drawings and specific embodiments.

[0055] As Figure 1-2 shown, an embodiment of the present invention provides an adaptive garbage cleaning method based on neuro-fuzzy, including the following steps:

[0056] S1. Construct a multi-modal garbage image data set by annotating garbage types and dividing the training set, validation set, and test set;

[0057] S2. Train an object detection model based on multi-modal feature fusion, fuse fuzzy clustering features and deep learning features, and dynamically optimize the classification boundary;

[0058] S3. Blur the garbage image, generate a membership matrix by combining the fuzzy clustering algorithm, and set a threshold to determine the garbage category;

[0059] S4. Establish a fuzzy rule base and achieve real-time decision-making through a fuzzy inference model;

[0060] S5. Train the model using a hybrid optimization strategy, combine global search and second-order accuracy tuning to improve the model efficiency and convergence speed;

[0061] S6. Based on 3D perception technology, estimate the volume and mass of garbage in real time, and match physical parameters by combining with a density database;

[0062] S7. Dynamically adjust the power of the cleaning device through a fuzzy PID controller to achieve the coordinated optimization of energy consumption and efficiency;

[0063] S8. Deploy an edge-cloud collaborative architecture to support real-time decision-making and incremental update of the rule base.

[0064] In this embodiment, step S1 is to construct a multi-modal garbage image data set by annotating garbage types and dividing the training set, validation set, and test set. The following is a detailed step description:

[0065] Step S1-1: Selection and deployment of multi-modal data acquisition equipment:

[0066] Select an RGB-D camera with excellent performance to ensure that it has high-resolution imaging ability and accurate depth information acquisition function, and can clearly capture the appearance features and spatial positions of garbage; at the same time, a lidar is configured, and its point cloud resolution needs to meet the requirement of accurately identifying different garbage shapes and distribution situations; these devices are reasonably installed on a movable acquisition platform (such as a vehicle, robot, etc.) to ensure that when collecting data in different scenarios (such as urban streets, community garbage dumps, parks, etc.), the device's perspective can fully cover the garbage area;

[0067] Step S1-2: Acquisition of original data:

[0068] Control the acquisition platform to move along a preset path in the selected garbage scenario. During this period, the RGB-D camera continuously captures garbage images at a set frame rate, synchronously obtains the corresponding depth image data, and the lidar scans the surrounding environment in real time to generate the point cloud data of the garbage; During the acquisition process, fully consider the impact of different times, weather and other factors on the appearance of garbage. For example, data is collected under sunny, cloudy, rainy days and different lighting conditions in the morning and evening to ensure the diversity of data;

[0069] Step S1-3: Preliminary screening of raw data:

[0070] After the acquisition is completed, conduct a preliminary screening of the large amount of raw data obtained; Eliminate images and point cloud data with incomplete data or seriously poor quality due to equipment failures, occlusions, etc.; For example, directly remove RGB images that are blurred, partially missing or have serious noise interference, as well as depth images with abnormal depth information from the dataset to reduce the amount of invalid data for subsequent processing;

[0071] Step S1-4: Data preprocessing:

[0072] Image preprocessing: Denoise the screened RGB images, and use algorithms such as Gaussian filtering to remove random noise in the images and smooth the image details; Then perform size normalization to uniformly adjust all images to the same size, such as [specific size value], to meet the requirements of subsequent model processing; For depth images, also perform denoising and smoothing operations, and register them with the RGB images to ensure accurate spatial correspondence between the two;

[0073] Point cloud data preprocessing: Filter the point cloud data generated by the lidar to remove outliers and noise points to improve the quality of the point cloud data; Through downsampling algorithms, reduce the density of the point cloud data without losing key information, reduce the amount of data, and facilitate subsequent processing; At the same time, fuse the point cloud data with the RGB-D image data to establish a unified coordinate system so that different modality data can be used synergistically;

[0074] Step S1-5: Garbage type annotation:

[0075] Organize professional annotators and use specialized image annotation tools to annotate the garbage types of the preprocessed multi-modal data; The annotators need to carefully observe the appearance characteristics of the garbage in the RGB images, and combine the spatial information provided by the depth images and point cloud data to accurately mark the position and category of each type of garbage in the images, such as common garbage categories like plastic bottles, aluminum cans, waste paper, fruit peels, etc.; During the annotation process, formulate detailed annotation specifications and review mechanisms to ensure the accuracy and consistency of the annotation; For complex scenarios or garbage with difficult-to-determine categories, annotate through multi-person discussions or expert guidance;

[0076] Step S1-6: Dataset Division:

[0077] Using the method of stratified sampling, divide the labeled multi-modal garbage image dataset according to the ratio of 70% as the training set, 15% as the validation set, and 15% as the test set; ensure that each subset contains various garbage categories and data samples under different scenarios and lighting conditions, so as to ensure that the model can fully learn the diversity characteristics of the data during the training, validation, and testing processes, and avoid performance deviation of the model caused by uneven data distribution; after the division, save the data and corresponding annotation information of the training set, validation set, and test set respectively to prepare for subsequent model training and evaluation.

[0078] In this embodiment, step S2 is the training of an object detection model based on multi-modal feature fusion, aiming to fuse fuzzy clustering features and deep learning features, dynamically optimize the classification boundary, and thus improve the accuracy of garbage recognition and the performance of the model; the detailed steps are as follows:

[0079] Step S2-1: Generate the membership matrix by FCM fuzzy clustering:

[0080] Determine the number of clusters : Combine factors such as the complexity of garbage types and the diversity of images in the multi-modal garbage image dataset to determine an appropriate number of clusters ; for example, if the dataset contains various garbage with different shapes and materials, a larger value may be required to better distinguish the characteristics of different types of garbage; if the types of garbage are relatively single, a smaller value can meet the requirements;

[0081] Run the FCM algorithm: Input the multi-modal garbage image data preprocessed in step S1 into the FCM fuzzy clustering algorithm; during the operation of the algorithm, it will update the membership degree of each sample (pixel points in the image or feature points after feature extraction, etc.) belonging to each cluster through continuous iteration according to the similarity measure between data points (such as Euclidean distance, etc.); in each iteration, the algorithm will adjust the cluster center and membership degree simultaneously, so that the objective function (such as the sum of squared errors) gradually decreases; when the change amount of the objective function is less than the preset threshold (such as 0.001) or the objective function value is basically stable after continuous multiple iterations, the algorithm converges and generates a stable membership matrix , where, is the membership matrix generated by FCM fuzzy clustering, is the number of clusters, is the number of samples; this matrix reflects the fuzzy membership relationship between each sample and each cluster, that is, each sample belongs to different clusters to different degrees;

[0082] Step S2-2: YOLOv8 outputs feature maps:

[0083] Model loading and configuration: Select the trained YOLOv8 model and load it onto a device with computing capabilities (such as a GPU); at the same time, according to the performance and resource conditions of the device, reasonably configure the running parameters of the model, such as setting the batch size, adjusting memory allocation, etc., to ensure that the model can run efficiently and stably;

[0084] Image input and feature extraction: Input multi-modal garbage images into the YOLOv8 model one by one in sequence; the model first performs convolutional operations on the input images, slides different convolutional kernels on the images to extract local features in the images; then performs pooling operations to downsample the feature maps obtained by convolution, reducing the amount of data while retaining key features; after multiple convolutional and pooling operations, feature maps containing rich semantic and spatial information are output at specific network layers of the model , where, is the feature map output by YOLOv8, is the height of the feature map, is the width of the feature map, is the number of channels of the feature map; these feature maps describe the features of the garbage from different scales and angles. For example, large-scale feature maps can capture the overall shape and position information of the garbage, while small-scale feature maps focus on features such as the detailed texture of the garbage;

[0085] Step S2-3: Channel concatenation operation:

[0086] Data dimension adaptation: Carefully check the membership matrix and the feature map output by YOLOv8; if their dimensions do not match, adjust according to the actual situation; for example, if does not meet the concatenation requirements in terms of dimensions, it may be necessary to perform dimension expansion on it and add axes with dimension 1 at appropriate dimensions; if some dimensions of the feature map are too large or too small, it may be necessary to perform cropping or padding operations to ensure that they can be correctly concatenated in the channel dimension;

[0087] Concatenation fusion: Concatenate the membership matrix with the feature map in the channel dimension; the new feature obtained after concatenation contains the fuzzy sample classification information provided by fuzzy clustering and the image feature information extracted by the deep learning model; for example, for a garbage image, the fuzzy clustering information can reflect the fuzzy association between different regions in the image and different garbage categories, while the deep learning feature map provides the intuitive visual features of the garbage in the image. The combination of the two can more comprehensively describe the features of the garbage and enhance the expression ability of the garbage features;

[0088] Step S2-4: The dynamic weight allocation module calculates the feature weights:

[0089] Set the temperature coefficient : Determine the value of the temperature coefficient based on experience and the results of multiple experiments ; Plays a key regulatory role in the smoothness of weight allocation; if the value is large, then the difference between the weights of different clustering features will be small, and the weight allocation will be more uniform, meaning that the importance of the features of each cluster in the fused features is relatively balanced; if the value is small, the weight difference will be more obvious, and the feature weight corresponding to the cluster with a high membership degree will be larger, and its feature will play a stronger leading role in the fused features; for example, in some experiments, it is found that when processing garbage images in complex backgrounds, a larger value can prevent the model from relying too much on a certain type of clustering feature and improve the robustness of the model; while in scenarios where the garbage categories are clearly distinguishable, a smaller value can better highlight the key clustering features and improve the classification accuracy;

[0090] Calculate the feature weights: According to the formula , calculate the corresponding feature weights for each cluster , where is the feature weight of the th cluster, is the membership degree of the th cluster, is the membership degree of the th cluster, is the temperature coefficient, is the number of clusters; during the calculation, the membership degrees and respectively represent the membership degrees of the th and th clusters, is the number of clusters; through this formula, the higher the membership degree of a cluster, the greater its corresponding feature weight, which also indicates that the features included in this cluster account for a larger proportion in the fused features and have a more significant impact on subsequent classification;

[0091] Step S2-5: Calculate the weighted fused features and input them into the detection head for classification:

[0092] Weighted summation: According to the feature weights calculated previously, perform a weighted summation operation on the cascaded features to obtain the weighted fused features ; where is the weighted fused feature, For the feature map corresponding to the th cluster; through weighted summation, the fused features can comprehensively consider the importance of different cluster features and effectively integrate the dominant features of each cluster;

[0093] Detector head classification: Input the weighted fused features into the detector head of the object detection model; the detector head usually contains a classifier (such as a Softmax classifier) inside. The classifier will calculate the probability that the garbage in the image belongs to each category according to the numerical distribution of the fused features; for example, the classifier will perform a series of operations on the fused features to obtain the probability values that the garbage belongs to different categories such as plastic bottles, aluminum cans, waste paper, etc., and finally select the category with the largest probability value as the predicted category label of the garbage, completing the training and classification process of the object detection model based on multi-modal feature fusion and achieving accurate identification of garbage categories.

[0094] In this embodiment, step S3 is to perform blurring processing on the garbage image, generate a membership matrix in combination with the fuzzy clustering algorithm, and set a threshold to determine the garbage category; the detailed steps are as follows:

[0095] Step S3-1: Select the fuzzy clustering algorithm: Select the FCM (Fuzzy C-Means Clustering) algorithm as the core algorithm of fuzzy clustering; the FCM algorithm realizes clustering of data by optimizing the objective function. It can divide data points into different clusters, and each data point has a membership degree to each cluster, which can well handle the fuzziness and uncertainty of data and is suitable for the classification processing of garbage images;

[0096] Step S3-2: Initialize parameters: Determine the value of the fuzzy index ; in the present invention, it is set as ; the fuzzy index affects the fuzziness of the clustering result, the larger it is, the more fuzzy the clustering result is, and the smaller the difference in the membership degree of each data point to different clusters is; the smaller it is, the closer the clustering result is to hard clustering, and the membership degree of the data point to a certain cluster tends to be 0 or 1; at the same time, determine the number of clusters , the number of clusters needs to be determined according to the approximate number of garbage categories in the garbage image dataset and the distribution characteristics of the data. For example, if there are mainly three types of garbage, plastic, metal, and paper, in the dataset, it can be initially set as , and it can be adjusted according to the experimental results later; in addition, it is also necessary to randomly initialize the cluster centers , and the initial values of the cluster centers will affect the convergence speed of the algorithm and the final clustering result;

[0097] Step S3-3: Calculate the membership matrix: For each sample in the garbage image ( , being the number of samples), according to the membership update formula of the FCM algorithm , where is the membership of the -th sample belonging to the -th cluster, is the -th sample, is the -th cluster center, represents the -th cluster center, is the fuzzy index, and calculate its membership belonging to the -th cluster ; during the calculation process, continuously update the membership matrix and the cluster center until the change of the membership matrix is less than a certain preset threshold (such as 0.001), indicating that the algorithm converges and a stable membership matrix is obtained;

[0098] Step S3-4: Set a threshold to determine the garbage category: When setting the membership threshold , it is determined as "large garbage"; traverse the membership matrix, and for each sample, if its membership to a certain cluster is greater than or equal to 0.7, then determine the garbage corresponding to the sample as "large garbage"; otherwise, further analyze or determine it as other categories according to the actual situation;

[0099] Step S3-5: Verify the classification accuracy: Use a confusion matrix to verify the classification accuracy; compare the garbage categories predicted by the model with the actual labeled garbage categories to construct a confusion matrix; for example, for two categories of "large garbage" and "non-large garbage", the confusion matrix will record the number of samples that are actually "large garbage" and are correctly predicted as "large garbage", the number of samples that are actually "large garbage" but are incorrectly predicted as "non-large garbage", the number of samples that are actually "non-large garbage" and are correctly predicted as "non-large garbage", and the number of samples that are actually "non-large garbage" but are incorrectly predicted as "large garbage"; by calculating the various indicators in the confusion matrix, the classification accuracy is obtained; the present invention requires the classification accuracy to be greater than 92%, if this accuracy is not reached, then the algorithm parameters (such as the number of clusters C, the fuzzy index m, etc.) need to be adjusted or the data preprocessing needs to be performed again, and then clustering and verification are performed again until the accuracy requirement is met.

[0100] In this embodiment, step S4 is to establish a fuzzy rule base and implement real-time decision-making through a fuzzy inference model to better control the garbage cleaning process; the detailed description of this step is as follows:

[0101] Step S4-1: Determine the input and output variables:

[0102] Input variable selection: Considering the key influencing factors in the actual scenario of garbage cleaning comprehensively, determine the input variables; these variables are usually related to the state of the garbage and the operation of the cleaning device; for example, the detection results of the current garbage (including information such as garbage category, quantity, distribution, etc.), which can reflect the complexity of the garbage cleaning task; the current power of the cleaning device, which is directly related to the cleaning efficiency and energy consumption; and the environmental parameters of the cleaning area (such as ground material, presence of obstacles, etc.), these factors will affect the cleaning difficulty.

[0103] Output variable determination: According to the control requirements of the garbage cleaning task, determine the output variables; the main output variables are the control instructions for the cleaning device, such as the power adjustment amount, the movement direction of the cleaning head, and the speed adjustment amount, etc.; these output variables will directly act on the cleaning device to achieve precise control of the garbage cleaning process.

[0104] Step S4-2: Define fuzzy linguistic variables:

[0105] Fuzzification of input variables: For each input variable, according to its value range and actual meaning, divide different fuzzy linguistic values; for example, for the input variable of garbage quantity, fuzzy linguistic values such as "extremely few", "few", "medium", "many", "extremely many" can be defined; for the current power of the cleaning device, define "low", "relatively low", "moderate", "relatively high", "high", etc.; at the same time, determine the corresponding membership functions for each fuzzy linguistic value, and common membership functions include Gaussian type, triangular, trapezoidal, etc.; taking the Gaussian membership function as an example, if the membership function of "many" for garbage quantity is defined as , it is necessary to determine the central value and the standard deviation , so that this function can accurately describe the degree to which the garbage quantity belongs to "many".

[0106] Fuzzification of output variables: Similarly, perform fuzzification processing on the output variables; for the power adjustment amount, fuzzy linguistic values such as "substantially decrease", "slightly decrease", "remain unchanged", "slightly increase", "substantially increase" can be defined, and determine their respective membership functions; in this way, convert the precise input and output numerical values into fuzzy linguistic variables, which is more in line with the fuzzy characteristics in actual decision-making.

[0107] Step S4-3: Formulate fuzzy rules:

[0108] Basis for rule generation: Fuzzy rules are formulated based on expert experience, actual operation data, and in-depth understanding of the garbage cleaning process. For example, if it is detected that the amount of garbage is "large" and the current power of the cleaning device is "low", from experience, to efficiently clean the garbage, the power of the cleaning device needs to be increased. Thus, a fuzzy rule is generated: "IF the amount of garbage is 'large' and the current power of the cleaning device is 'low', THEN the power adjustment amount is 'a large increase'".

[0109] Consideration of rule comprehensiveness: Try to cover all possible combinations of input variables as much as possible, formulate a series of complete fuzzy rules, and form a fuzzy rule base. The rules in the rule base should be coordinated with each other to avoid contradictions or conflicts, and ensure that reasonable decision-making suggestions can be given in different garbage cleaning scenarios.

[0110] Step S4-4: Construct a fuzzy inference model:

[0111] Select an inference method: Common fuzzy inference methods include Mamdani inference method and Sugeno inference method, etc. The present invention can select a suitable inference method according to actual needs. Taking the Mamdani inference method as an example, it performs inference based on the composition operation of fuzzy relations.

[0112] Implementation of the inference process: When there is new input data, first, according to the membership functions of the input variables, determine the membership degrees of the input data to each fuzzy linguistic value. Then, based on the fuzzy rule base, find the matching fuzzy rules. For each matching rule, determine the activation strength of the rule according to the input membership degrees (usually taking the minimum value of each input membership degree). Finally, synthesize the outputs of all activated rules to obtain the final fuzzy output result.

[0113] Step S4-5: Defuzzification:

[0114] Method selection: Since the result obtained from fuzzy inference is a fuzzy set, a defuzzification operation needs to be performed to convert the fuzzy output into an exact control quantity. Common defuzzification methods include the centroid method, the maximum membership degree method, etc. If the centroid method is adopted, the calculation formula is , where is an element in the domain of the output variable, is its corresponding membership degree.

[0115] Output exact control instructions: Through defuzzification calculation, obtain exact output values, such as specific power adjustment values, cleaning head movement speed adjustment values, etc. These exact control instructions will be transmitted to the actuator of the cleaning device to achieve real-time control of the garbage cleaning device to adapt to different garbage cleaning scenarios.

[0116] In this embodiment, step S5 trains the model using a hybrid optimization strategy, combining global search and second-order accuracy tuning to improve the model efficiency and convergence speed. The specific steps are as follows:

[0117] Step S5-1: Grid search stage:

[0118] Determine the parameter space: Define the value range of the parameters to be optimized and construct the parameter space

[0119] ; where and are important parameters closely related to the model performance, and their values will have a significant impact on the training effect of the model;

[0120] Traverse the parameter combinations: Within the established parameter space, traverse all possible combinations of and ; this means that takes values within the range of with a certain step size, takes values within the range of with the corresponding step size, and then perform subsequent model training and evaluation on each group of combinations;

[0121] Screen the candidate parameters: For each group of parameter combinations, train the model using the training set data and evaluate the performance of the model through the validation set data; Using the error as the evaluation metric, screen out the parameter combinations with an error lower than the threshold as the candidate parameters for further optimization; These candidate parameters represent the parameter settings that can enable the model to meet certain performance requirements within the current parameter space search range;

[0122] Step S5-2: L-M optimization stage:

[0123] Initialize based on the candidate parameters: Start the L-M (Levenberg-Marquardt) optimization process with the candidate parameters screened in the grid search stage as the initial values; L-M optimization is an optimization algorithm for nonlinear least squares problems, especially suitable for fine-tuning model parameters;

[0124] Calculate the relevant matrix and vector: During the optimization process, it is necessary to calculate the Jacobian matrix of the error with respect to the parameters; The Jacobian matrix describes the rate of change of the error function with respect to each parameter, and it reflects the sensitivity of the model error to parameter changes; At the same time, determine the damping factor , whose initial value is set to , and the dynamic adjustment range is ; In addition, it is also necessary to calculate the error vector , which represents the difference between the model predicted value and the true value;

[0125] Parameter update iteration: According to the update formula optimized by L-M , calculate the update amount of the parameter , where is the update amount of the parameter, is the Jacobian matrix of the error with respect to the parameter, is the Jacobian matrix of the transposed matrix, is the damping factor, is the identity matrix, is the error vector; By continuously iterating to calculate the update amount and update the parameters, the model parameters are gradually adjusted to make the model error gradually decrease until the preset convergence condition is met (such as the parameter update amount is less than a certain threshold, or the model error no longer significantly decreases in consecutive multiple iterations); In each iteration process, the damping factor is dynamically adjusted according to the performance of the model ; When the model converges quickly, appropriately reduce to accelerate the convergence speed; When the model is unstable or oscillates, increase to ensure the stability of the algorithm; After multiple rounds of iterative optimization, a set of parameters optimized with second-order accuracy is obtained, and these parameters can effectively improve the efficiency and convergence speed of the model, making the model perform better in the garbage cleaning task.

[0126] In this embodiment, step S6 estimates the garbage volume and mass in real time based on three-dimensional perception technology, combines with the density database to match physical parameters, and provides more accurate data support for garbage cleaning work. The specific steps are as follows:

[0127] Step S6-1: Obtain relevant parameters and data:

[0128] Determine the physical size of the pixel: According to the physical width of the sensor and the width of the image , calculate the physical size of the pixel in the direction; Similarly, based on the physical height of the sensor and the height of the image , obtain the physical size of the pixel in the direction, where is the physical height of the sensor, is the height of the image; These physical size parameters are the basis for accurately estimating the garbage volume later;

[0129] Collect depth value data: Use three-dimensional perception devices (such as RGB-D cameras, lidar, etc.) to obtain the depth value of each pixel in the garbage image ( , (where \(N\) is the number of pixels); the depth value reflects the distance information between the garbage and the sensor, and accurate acquisition of the depth value is crucial for accurately estimating the volume of the garbage;

[0130] Step S6-2: Calculate the three-dimensional volume of the garbage:

[0131] Execute the volume estimation formula: According to the formula , perform cumulative calculation on the tiny volume elements corresponding to each pixel; during the calculation process, multiply the physical size of each pixel in the , direction by the corresponding depth value to obtain the tiny volume represented by each pixel, and then add up the tiny volumes of all pixels to obtain the three-dimensional volume of the garbage ; this calculation process takes into account the actual occupancy of the garbage in three-dimensional space and can more accurately estimate the volume of the garbage;

[0132] Step S6-3: Match the garbage density and calculate the mass:

[0133] Query the density database: From the preset database containing density values of 20 types of garbage, match the corresponding garbage density according to the previously detected garbage category ; for example, if the detected garbage is a plastic bottle, find the density value corresponding to the plastic bottle in the database;

[0134] Calculate the mass of the garbage: Use the formula , multiply the matched garbage density by the calculated garbage volume to obtain the mass of the garbage ; in this way, by combining the volume and density information of the garbage, accurate estimation of the garbage mass is achieved, providing important reference data for power adjustment, transportation planning, etc. of subsequent garbage cleaning equipment.

[0135] In this embodiment, step S7 aims to dynamically adjust the power of the cleaning device through a fuzzy PID controller to achieve coordinated optimization of energy consumption and efficiency. The detailed steps are as follows:

[0136] Step S7-1: Determine the input and output of the fuzzy PID controller:

[0137] Clarify the input variables: Select the error at the current moment and the rate of change of the error as the input variables of the fuzzy PID controller; among them, the error is calculated by comparing the actual working state of the current cleaning device (such as the difference between the garbage cleaning amount, cleaning area coverage, etc. and the expected target); the rate of change of the error It is obtained by calculating the change of the error at adjacent moments, reflecting the change trend of the error;

[0138] Determine the output variable: Take the adjustment amount of the power as the output variable of the fuzzy PID controller; this variable directly acts on the cleaning device to adjust the power of the cleaning device in real time to adapt to different garbage cleaning scenarios;

[0139] Step S7-2: Define fuzzy language variables and membership functions:

[0140] Divide the fuzzy language values: Divide the fuzzy language values for the input variables and ; for example, divide into "large positive", "medium", "small negative", etc.; divide into "positive fast", "zero", "negative slow", etc.; these fuzzy language values can more intuitively describe the state of the error and its change rate;

[0141] Determine the membership function: Use the Gaussian membership function to quantify the fuzzy language values; taking the membership function of the error belonging to "large" as an example, it is defined as , where is the membership of the error belonging to "large", is the center of the Gaussian membership function, is the standard deviation of the Gaussian membership function, , is the center of the Gaussian membership function, , is the standard deviation, response time < 200ms; in this way, the precise input numerical values are converted into fuzzy language variables, better reflecting the fuzziness in actual control;

[0142] Step S7-3: Construct a fuzzy rule base:

[0143] Formulate fuzzy rules: Formulate fuzzy rules based on expert experience and actual control requirements; for example:

[0144] Rule 1: If is "large positive" and is "positive fast", THEN , where is the error at the current moment; is the change rate of the error; is the adjustment amount of the power; is the proportional coefficient; is the differential coefficient, this rule is applicable to the situation where the error is large and the error is increasing rapidly, and the power is quickly adjusted through proportional and differential control to reduce the error;

[0145] Rule 2: If is "medium" and is "zero", THEN , when the error is at a medium level and the change rate is zero, adjust the power only through proportional control;

[0146] Rule 3: If is "small negative" and is "negative slow", , where is the integral coefficient; is the integral of the error. When the error is small and decreasing slowly, use integral control to eliminate the steady-state error and further optimize the power adjustment;

[0147] Improve the rule base: Ensure that the fuzzy rule base covers all possible input combination cases, enabling the controller to make reasonable decisions under different working conditions; The rules should be coordinated with each other to avoid contradictions or conflicts;

[0148] Step S7-4: Fuzzy inference and defuzzification:

[0149] Fuzzy inference process: When there are new input data and , determine their membership degrees to each fuzzy linguistic value according to the membership function; Then, find the matching fuzzy rules based on the fuzzy rule base and determine the activation strength of each rule according to the input membership degrees (usually taking the minimum of the input membership degrees); Finally, synthesize the outputs of all activated rules to obtain the fuzzy output result;

[0150] Defuzzification operation: Use a suitable defuzzification method (such as the centroid method, etc.) to convert the fuzzy output into an accurate power adjustment amount ; For example, the centroid method obtains the accurate value by calculating the centroid of the fuzzy output set, and the formula is ;

[0151] Step S7-5: Power regulation and effect monitoring:

[0152] Adjust the power of the cleaning device: Apply the power adjustment amount obtained by defuzzification to the cleaning device and adjust its power in real time; For example, when is positive, increase the power of the cleaning device; when is negative, reduce the power of the cleaning device;

[0153] Monitoring and feedback: After power adjustment, continuously monitor the working state of the cleaning device, including indicators such as garbage cleaning efficiency and energy consumption; According to the monitoring results, continuously adjust the parameters of the fuzzy PID controller (such as , , ), to further optimize the power regulation effect and achieve the coordinated optimization of energy consumption and efficiency; at the same time, ensure that the response time of the entire regulation process is less than 200 ms to meet the requirements of real-time control.

[0154] Step S8 deploys an edge-cloud collaborative architecture to support real-time decision-making and incremental update of the rule base. The specific steps are as follows:

[0155] Step S8-1: Deploy a lightweight TSK fuzzy inference model at the edge side:

[0156] Model selection and configuration: Select a lightweight TSK (Takagi-Sugeno-Kang) fuzzy inference model and deploy it on the edge device; this model has a relatively simple structure and low computational complexity, making it suitable for running on resource-constrained edge devices; according to the hardware performance of the edge device (such as computing power and memory size) and the requirements of the garbage cleaning task, configure the model to determine the number of input variables , and initialize the rule parameters of the model, that is, the constant term of the th rule and the coefficient of the th rule for the th input variable; ;

[0157] Data input and inference: The edge device obtains real-time data such as images and depth information collected by the multi-modal data acquisition unit and uses it as an input vector to input into the lightweight TSK fuzzy inference model; the model calculates the output of each rule according to the preset rules ; through operations such as weighted summation of all rule outputs, the final inference result is obtained to support real-time decision-making, such as controlling the actions of the cleaning device;

[0158] Incremental learning to update parameters: During the garbage cleaning process, as new data is continuously obtained, the edge side uses an incremental learning algorithm to update the rule parameters of the TSK fuzzy inference model; for example, when encountering a new garbage distribution scenario or a change in garbage category characteristics, the model adjusts the constant term and the coefficient to enable the model to adapt to environmental changes and improve the accuracy of decision-making;

[0159] Step S8-2: Execute density database expansion and synchronization in the cloud:

[0160] Data collection and analysis: The cloud collects the garbage data uploaded from each edge device every month, including garbage images, volume and mass estimation data, and information on garbage samples that are difficult to match the existing density values during the cleaning process, etc.; analyze this data to identify possible new garbage categories or changes in the density values of existing garbage categories;

[0161] Database expansion: According to the data analysis results, add the density values of the newly added garbage categories to the density database; for example, if a new type of plastic garbage is discovered, after determining its density value through experimental measurement or referring to relevant materials, enter this value into the database to enrich the content of the database and provide more accurate data support for subsequent garbage mass estimation;

[0162] Synchronize to edge devices: Use OTA (Over-The-Air) technology to synchronize the updated density database to each edge device; after the edge device receives the update information, it automatically downloads and updates the local density database to ensure that the edge side can use the latest density data when estimating the garbage mass based on the three-dimensional perception technology, improving the accuracy of garbage mass estimation.

[0163] Please refer to Figure 2 , a neural fuzzy-based adaptive garbage cleaning system, including:

[0164] Multimodal data acquisition unit: Integrate RGB-D cameras and lidar to collect images and depth information;

[0165] Hybrid computing unit: Deploy a neural fuzzy-based adaptive garbage cleaning method, including an FPGA-accelerated fuzzy inference module. The FPGA-accelerated fuzzy inference module adopts a parallel architecture to perform hardware acceleration on the fuzzy inference process, with a single-frame processing delay < 50ms and a power consumption < 5W;

[0166] Dynamic power control unit: Support power adjustment from 50W to 150W, and the suction non-linear mapping formula is , where is the suction force, is the power, is the coefficient, is the exponent, , , and the maximum suction force ≥ 200N;

[0167] Cloud management platform: Provide rule library version control and incremental learning interfaces;

[0168] Furthermore, this garbage cleaning system integrates functions such as multimodal data acquisition, hybrid computing, dynamic power control, and cloud management, and realizes intelligent garbage cleaning through multi-technology collaboration, improving the cleaning efficiency, reducing energy consumption, and adapting to complex scenarios. Specifically:

[0169] Multi-modal data acquisition unit: Integrates an RGB-D camera and a lidar, capable of simultaneously acquiring image and depth information; The RGB-D camera is responsible for taking color images of the garbage, providing appearance features of the garbage, such as color, shape, texture, etc.; The lidar generates point cloud data of the garbage by emitting laser beams and receiving reflected signals, accurately obtaining the spatial position and depth information of the garbage; The two work together to comprehensively collect multi-modal data of the garbage, providing rich data support for subsequent garbage recognition, volume estimation, etc.; In different scenarios, such as urban streets, community garbage dumps, etc., it can stably collect data, and considering the impact of different times and weather on the appearance of the garbage, it ensures the diversity and comprehensiveness of the collected data;

[0170] Hybrid computing unit: Deploys an adaptive garbage cleaning method based on neuro-fuzzy, with the core being a fuzzy inference module accelerated by FPGA; This module adopts a parallel architecture and utilizes the hardware parallel characteristics of FPGA to accelerate the fuzzy inference process in hardware; Compared with the traditional software computing method, it greatly improves the computing speed, with a single-frame processing delay less than 50ms and a power consumption lower than 5W; When processing garbage images, it can quickly fuse fuzzy clustering features and deep learning features, dynamically optimize the classification boundary, and accurately identify garbage categories; By establishing a fuzzy rule base, it realizes real-time decision-making, efficiently processes a large amount of data, provides precise control instructions for garbage cleaning, and improves the operating efficiency of the entire system;

[0171] Dynamic power control unit: Supports power adjustment from 50W to 150W, and dynamically adjusts the power of the cleaning device according to the output of the fuzzy PID controller; Through the suction non-linear mapping formula ( , )), it realizes the precise matching of power and suction, and the maximum suction is not less than 200N; During the garbage cleaning process, it dynamically adjusts the power according to the volume, mass, distribution of the garbage and the real-time working state of the cleaning device; When encountering large-volume or heavy garbage, it increases the power to improve the suction; When the garbage is less or lighter, it reduces the power to save energy consumption, thereby realizing the coordinated optimization of energy consumption and efficiency;

[0172] Cloud management platform: Provides rule library version control and incremental learning interfaces; On the one hand, it manages the version of the fuzzy rule library, records the content and time of each update of the rule library, facilitating traceability and management; On the other hand, it supports incremental learning. When the edge device uploads new garbage data or discovers new garbage categories, the cloud uses this data for learning and updates the rule library; It also performs density database expansion every month, adding density values of new garbage categories and synchronizing them to the edge device through OTA, enabling the entire system to continuously adapt to new garbage types and scenarios, and continuously improving the accuracy and adaptability of garbage cleaning.

[0173] Although embodiments of the present invention have been shown and described, those of ordinary skill in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An adaptive garbage cleaning method based on neuro-fuzzy, characterized in that: It includes the following steps: S1. Construct a multi-modal garbage image dataset by annotating garbage types and dividing the training set, validation set, and test set; S2. Train an object detection model based on multi-modal feature fusion, fuse fuzzy clustering features and deep learning features, and dynamically optimize the classification boundary. The multi-modal feature fusion specifically includes: The membership matrix generated by FCM fuzzy clustering and the feature map output by YOLOv8 are cascaded at the channel level, where is the membership matrix generated by FCM fuzzy clustering, is the number of clusters, is the number of samples, is the feature map output by YOLOv8, is the height of the feature map, is the width of the feature map, is the number of channels of the feature map, and the feature weights are calculated through the dynamic weight allocation module. The formula is: , where is the feature weight of the th cluster, is the membership degree of the th cluster, is the membership degree of the th cluster, is the temperature coefficient, is the number of clusters; Weighted fusion features Input to the detection head for classification, where is the weighted fusion feature, is the feature map corresponding to the th cluster; S3. Blur the garbage images, generate a membership matrix by combining the fuzzy clustering algorithm, and set a threshold to determine the garbage category; S4. Establish a fuzzy rule base and achieve real-time decision-making through a fuzzy inference model; S5. Train the model using a hybrid optimization strategy, combine global search and second-order accuracy tuning to improve the model efficiency and convergence speed; S6. Estimate the volume and mass of garbage in real time based on 3D perception technology, and match physical parameters by combining with a density database; S7. Dynamically adjust the power of the cleaning device through a fuzzy PID controller to achieve the collaborative optimization of energy consumption and efficiency; S8. Deploy an edge-cloud collaborative architecture to support real-time decision-making and incremental update of the rule base.

2. The adaptive garbage cleaning method based on neuro-fuzzy according to claim 1, characterized in that: The fuzzy clustering algorithm in step S3 is FCM, and its membership update formula is: , where is the membership degree that the -th sample belongs to the -th cluster, is the -th sample, is the -th cluster center, represents the -th cluster center, is the fuzzy index; Settings , membership degree When it is judged as "big garbage", and the classification accuracy is verified to be >92% through the confusion matrix.

3. An adaptive garbage cleaning method based on neuro-fuzzy according to claim 1, characterized in that: The hybrid optimization strategy in step S5 includes: Grid search stage: traverse combinations within the parameter space to screen out candidate parameters with errors lower than the threshold , where and are parameters to be optimized is the error threshold; L-M optimization stage: Perform second-order optimization based on candidate parameters, and the update formula is: , where is the update amount of the parameter, is the Jacobian matrix of the error with respect to the parameter, is the Jacobian matrix of the transpose matrix, is the damping factor, is the identity matrix, is the error vector, and the initial value of the damping factor , and the dynamic adjustment range is .

4. A neural-fuzzy based adaptive garbage cleaning method according to claim 1, characterized in that: In step S6, the 3D volume estimation formula is: , where is the three-dimensional volume of the garbage; is the number of pixels; is the physical size of the pixel in the direction, , is the physical size of the pixel in the direction, , is the physical width of the sensor, is the width of the image, is the physical height of the sensor, is the height of the image; is the depth value of the th pixel; The quality is calculated as , is the mass of the garbage, is the density of the garbage, and the density is matched from a preset database, and the database contains 20 types of garbage density values.

5. The adaptive garbage cleaning method based on neuro-fuzzy according to claim 1, characterized in that: In step S7, the rule base of the fuzzy PID controller includes: Rule 1: is "Taisho" and is "Zhengkuai", THEN , where is the error at the current moment; is the rate of change of the error; is the adjustment amount of the power; is the proportionality coefficient; is the differential coefficient; Rule 2: is "medium" and is "zero", THEN ; Rule 3: is "small negative" and is "negative slow", , where is the integral coefficient; is the integral of the error; The membership function is of Gaussian type and is defined as , where is the error vector belongs to the membership degree of "large", is the center of the Gaussian membership function, is the standard deviation of the Gaussian membership function, , , and the response time < 200 ms.

6. The adaptive garbage cleaning method based on neural fuzzy according to claim 1, characterized in that: The edge-cloud collaborative architecture in step S8 includes: Deploy a lightweight TSK fuzzy inference model at the edge side, and its rule output is: , where is the output of the th rule, is the input vector; is the constant term of the th rule; is the th rule, and the th coefficient of the input variable, updated by incremental learning, is the number of input variables; The cloud performs density database expansion monthly, adding new density values for garbage categories and synchronizes them to edge devices via OTA.

7. An adaptive garbage cleaning system based on neuro-fuzzy, characterized in that: It includes: Multi-modal data acquisition unit: Integrate an RGB-D camera and a lidar to acquire images and depth information; Hybrid computing unit: Deploy the method described in any one of claims 1-6, including an FPGA-accelerated fuzzy inference module. The FPGA-accelerated fuzzy inference module adopts a parallel architecture to perform hardware acceleration on the fuzzy inference process, with a single-frame processing delay <50ms and a power consumption <5W; Dynamic power control unit: Supports power adjustment from 50W to 150W, and the suction non-linear mapping formula is , where is the suction force, is the power, is the coefficient, is the exponent, , , and the maximum suction force ≥ 200N; Cloud management platform: Provide rule base version control and incremental learning interfaces.

Images