An intelligent supervision system for garbage classification and placement

Through the intelligent supervision system, the garbage disposal errors are identified and corrected in real time, and the neural network model of face recognition and attention mechanism is used to solve the problem that the garbage classification disposal system cannot correct in real time and residents lack knowledge, and improve the garbage classification accuracy and garbage classification awareness .

CN118552060BActive Publication Date: 2024-11-26SUZHOU LECHUANG ENVIRONMENTAL PROTECTION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411012132.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-26
Publication Date
2024-11-26
Estimated Expiration
2044-07-26

AI Technical Summary

Technical Problem

The existing garbage classification and disposal system cannot correct the delivery error in real time, and residents lack garbage classification knowledge, resulting in the failure to dispose of garbage correctly.

Method used

An intelligent supervision system was designed, including information entry module, information verification module, garbage recognition module, recognition result judgment module, voice prompt module, classification analysis module, supervision and evaluation module and human-computer interaction module, using face recognition and attention The force mechanism neural network model performs garbage identification and classification analysis, corrects errors in real time and improves the accuracy of garbage disposal.

Benefits of technology

By identifying and correcting garbage disposal errors in real time, the accuracy of garbage classification has been improved, residents' awareness of garbage classification and knowledge popularization have been enhanced, and the overall effect of garbage classification has been improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118552060B_ABST
    Figure CN118552060B_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of garbage placement, and discloses an intelligent supervision system for garbage classification placement, comprising an information input module, an information verification module, a garbage identification module, an identification result judgment module, a voice prompt module, a classification situation analysis module, a supervision evaluation module and a human-computer interaction module. By providing the garbage identification module, the classification situation analysis module and the supervision evaluation module, it is beneficial to identify the input garbage based on a neural network model, and introduce an attention mechanism into the neural network module to increase the weight of channels with functional characteristics and suppress ineffective channels, so as to obtain accurate garbage identification results and assist placement personnel to place the garbage into the correct garbage bin; based on the placement accuracy of each category of garbage and the proportion of incorrectly placed garbage categories, the garbage classification awareness of personnel and the popularization effect of garbage classification knowledge are obtained, thereby providing a guiding direction for subsequent supervision work.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of garbage disposal, and more particularly to an intelligent supervision system for classified garbage disposal. Background Art

[0002] In order to improve the quality of garbage classification, behavioral supervision and on-site management of garbage classification are both basic and important tasks in the current garbage classification work. With the development of artificial intelligence, an intelligent garbage classification supervision system has been proposed. That is, first a certain identity authentication is performed on the individual who puts out the garbage, and then the garbage put into the garbage recycling system is classified and identified through image recognition, and the identification result information is recorded in the background system to facilitate the subsequent supervision and correction of the behavior of garbage classification.

[0003] However, the subsequent supervision of the garbage classification and disposal behavior makes it impossible for the relevant supervisors to obtain the real-time disposal situation, so they cannot correct the disposal behavior in a timely manner and cannot urge residents to sort the incorrectly disposed garbage to the correct disposal location; at the same time, the garbage classification and disposal is sometimes not done intentionally by residents, but due to a lack of knowledge about garbage classification, they are unable to put the garbage into the corresponding area correctly. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides an intelligent supervision system for garbage classification and disposal to solve the problems existing in the above-mentioned background technology.

[0005] The present invention provides the following technical solution: an intelligent supervision system for garbage classification and placement, comprising an information entry module, an information verification module, a garbage identification module, an identification result judgment module, a voice prompt module, a classification situation analysis module, a supervision evaluation module, and a human-computer interaction module;

[0006] The information entry module is used to enter personnel information and store it in the database, and to label the entered personnel, sequentially as U1, U2, U3...UN, where N is the total number of entered personnel;

[0007] The information verification module is used to verify the information of the delivery personnel, and verify the information of the delivery personnel through face recognition technology, and at the same time transmit the personnel information and the input personnel number to the supervision and evaluation module;

[0008] The garbage recognition module recognizes the input garbage based on a neural network model that introduces an attention mechanism, and outputs the recognition results to the recognition result discrimination module;

[0009] The recognition result discrimination module is used to classify and discriminate the recognition result, that is, to compare and discriminate the garbage category corresponding to the recognition result with the category of the garbage bin, and transmit the discrimination result to the voice prompt module; the discrimination result includes two types: "correct classification" and "incorrect classification". If the discrimination result is "incorrect classification", the image of the garbage bin is transmitted to the classification situation analysis module at the same time;

[0010] The classification analysis module is used to receive data from the recognition result judgment module, store the images of garbage that are put into the wrong place, and count the types of garbage that are put into the wrong place to obtain the accuracy rate of each type of garbage.

[0011] The voice prompt module is used to receive data from the recognition result judgment module and make corresponding voice announcements according to the recognition results;

[0012] The supervision and evaluation module is used to supervise the personnel's correction of garbage placement. If the personnel does not change the wrong garbage placement category, the personnel information is recorded, the garbage placement accuracy rate of the input personnel and the overall garbage placement accuracy rate are calculated, and corresponding supervision operations are performed;

[0013] The human-computer interaction module is used to perform human-computer interaction display on the data.

[0014] Preferably, the garbage identification module identifies the input garbage based on the neural network model, including the following steps:

[0015] Step S01: Take the image of the garbage and mark it, which is marked as U corresponding to the number of the person who put it in. da , where d = 1, 2, 3...N, a = 1, 2, 3...E, E is the total number of garbage disposals, U da The junk image is placed by the dth person for the ath time, and then the image is processed;

[0016] Step S02: training the neural network model and introducing an attention mechanism; the neural network model is a residual network, including a convolutional layer, a pooling layer, a residual block structure, a fully connected layer, and an output layer;

[0017] Step S03: Based on the trained residual network, the junk image in step S01 is identified and the identification result is output.

[0018] Preferably, the classification analysis module calculates the accuracy of each type of garbage placement using the following formula:

[0019] , where β tt is the accuracy rate of garbage placement in the tth category, φ tt is the amount of garbage of the tth category put into the tth trash bin, γt is the total amount of garbage put into the tth garbage bin; if there are T garbage categories, corresponding to T garbage bins, then t=1, 2, 3...T;

[0020] Calculate the percentage of incorrectly placed garbage in each category of trash bins using the following formula: , where Z qt is the placement rate of the qth garbage category in the garbage bin corresponding to the tth garbage category, φ qt is the amount of garbage of the qth category put into the tth garbage bin, γ t The amount of all garbage put into the t-th garbage bin, and q≠t, q=1, 2, 3...T.

[0021] Preferably, the calculation formula of the garbage placement accuracy of the input personnel by the supervision and evaluation module is: , where λ d is the garbage delivery accuracy of the dth person, µ d is the correct number of garbage disposals by the dth person, E d The total number of times the dth person puts out garbage; d = 1, 2, 3...N;

[0022] The calculation formula for the overall garbage placement accuracy is: , where λ´ is the overall garbage placement accuracy.

[0023] Preferably, the convolution layer can obtain a feature image after convolution processing. Multiple feature maps constitute a convolution layer. All neurons in each feature map share the same convolution kernel parameters. The feature maps obtained by using different convolution algorithms are also different. If the convolution kernel k is used to convolve the input data to obtain a feature map, the formula is expressed as: ,in, is the j-th feature map of the l-th layer; is the output value of the previous layer, i.e. the l-1 layer; is the bias value; M j is the area where the convolution kernel acts on the image; Represents a convolution kernel; f represents a nonlinear activation function;

[0024] The size of the output feature map of each layer of convolution depends on the padding value, size, stride, and the size of the input image of the layer. The size of the output feature map of the convolution is expressed as: , where S output is the output feature map size of the convolution, S input is the size of the input image, F is the size of the convolution kernel, P is the number of padded pixels, and B is the sliding step size of the convolution kernel.

[0025] Preferably, the formula of the pooling layer is expressed as: ,in, is the input feature of the lth layer, and the output feature after pooling is obtained through pooling operation ; The pooling operation requires the pooling kernel to move along the length and width on the feature map of each channel. The feature map S output The size of is determined by the hyperparameters of the pooling operation, and the size of the feature map is expressed as follows: , where D output D is the size of the output feature map of the pooling layer. input is the length or width of the input feature map, P is the pixel to be filled, and D filter is the length or width of the pooling kernel, B filter is the moving step of the pooling kernel.

[0026] Preferably, the formula of the residual block structure is: H(x)=F(x)+x, where x is the input data, F(x) is the residual function, which is the real learning part of the model; H(x) is the output result. When F(x)=0, H(x)=x, which is the so-called identity mapping. The goal of the training process is to make the residual result as close to 0 as possible.

[0027] The fully connected layer combines the features extracted by the pooling layer and the convolutional layer for classification; the function expression of the fully connected layer is: , where z i is the output value of the i-th node, k is the number of corresponding output nodes, that is, the total number of corresponding classification types.

[0028] Preferably, the attention mechanism is an SE attention mechanism, which performs a pooling operation on the input feature map, then inputs it into the fully connected layer, uses the ReLU function to input it into the next fully connected layer learning, and then outputs it through the Sigmoid function, which is recorded as S'; the formula is expressed as: , where δ is the activation function ReLU, σ is the activation function Sigmoid, W1 and W2 are the weight matrices of the two fully connected layers respectively, , where Y is the number of hidden nodes in the middle layer, c is the number of image channels, , where W and H are the length and width of the image. Each layer of the input feature map undergoes a 2D spatial kernel convolution, and finally c output feature images are obtained to form the feature map u c。

[0029] Technical effects and advantages of the present invention:

[0030] The present invention is conducive to identifying the input garbage based on the neural network model by providing a garbage identification module, a classification analysis module and a supervision and evaluation module, and introduces an attention mechanism in the neural network module to ensure the importance of the network in the channel, increase the weight of the channel with effective characteristics, and suppress the ineffective channel, so as to obtain accurate garbage identification results and assist the garbage disposal personnel to put the garbage into the correct garbage bin; at the same time, the garbage classification of the personnel is comprehensively analyzed and supervised, and the accuracy rate of each category of garbage and the proportion of incorrectly placed garbage categories are obtained, so as to obtain the personnel's garbage classification awareness and the popularization effect of garbage classification knowledge, thereby providing a guiding direction for subsequent supervision work; and according to the values ​​of the input personnel's garbage disposal accuracy rate and the overall garbage disposal accuracy rate, corresponding supervision operations are performed to enhance the personnel's garbage classification awareness and the popularity of garbage classification, and fundamentally improve the garbage classification effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is a flow chart of the intelligent supervision system for garbage classification and disposal according to the present invention. DETAILED DESCRIPTION

[0032] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the drawings in the present invention. In addition, the forms of the various structures recorded in the following embodiments are merely examples. The intelligent supervision system for garbage classification and disposal involved in the present invention is not limited to the various structures recorded in the following embodiments. All other implementations obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0033] The present invention provides an intelligent supervision system for garbage classification and placement, including an information entry module, an information verification module, a garbage identification module, an identification result judgment module, a voice prompt module, a classification situation analysis module, a supervision evaluation module, and a human-computer interaction module;

[0034] The information entry module is used to enter personnel information and store it in the database, and to label the entered personnel in sequence as U1, U2, U3...UN, where N is the total number of entered personnel; the personnel information includes resident facial information and house number information;

[0035] The information verification module is used to verify the information of the delivery personnel. It verifies the information of the delivery personnel through face recognition technology. If the delivery personnel information is the same as the input personnel information stored in the database, the verification is passed. Otherwise, the verification fails. After the verification is passed, the trash can is automatically opened, and the personnel information and the input personnel number are transmitted to the supervision and evaluation module.

[0036] The garbage recognition module recognizes the input garbage based on a neural network model that introduces an attention mechanism, and outputs the recognition results to the recognition result discrimination module; the recognition results are the categories of garbage, including but not limited to recyclable garbage, non-recyclable garbage and kitchen waste;

[0037] The recognition result discrimination module is used to classify and discriminate the recognition result, that is, to compare and discriminate the garbage category corresponding to the recognition result with the category of the garbage bin, and transmit the discrimination result to the voice prompt module; the discrimination result includes two types: "correct classification" and "incorrect classification". If the discrimination result is "incorrect classification", the image of the garbage bin is transmitted to the classification situation analysis module at the same time;

[0038] The classification analysis module is used to receive data from the recognition result judgment module, store the images of garbage that are put into the wrong place, and count the types of garbage that are put into the wrong place to obtain the accuracy rate of each type of garbage.

[0039] The voice prompt module is used to receive data from the recognition result discrimination module and make corresponding voice announcements according to the recognition result; if the recognition result discrimination module determines that the classification is correct, the voice announcement module will announce that the garbage is correctly placed; if the recognition result discrimination module determines that the classification is incorrect, the voice announcement module will announce that the garbage is incorrectly placed and the correct garbage placement category will be announced;

[0040] The supervision and evaluation module is used to supervise the personnel's correction of garbage disposal. If the personnel does not change the incorrect garbage disposal category, the personnel information is recorded, the garbage disposal accuracy rate of the input personnel and the overall garbage disposal accuracy rate are calculated, and corresponding supervision operations are performed; the supervision operations include persuading personnel to dispose of garbage, conducting relevant garbage disposal lectures, and promoting garbage disposal knowledge;

[0041] The human-computer interaction module is used to perform human-computer interaction display on the data.

[0042] In this embodiment, it should be specifically explained that the garbage identification module identifies the input garbage based on the neural network model, including the following steps:

[0043] Step S01: Take the image of the garbage and mark it, which is marked as U corresponding to the number of the person who put it in. da , where d = 1, 2, 3...N, a = 1, 2, 3...E, E is the total number of garbage disposals, U da The garbage image is placed by the dth person for the ath time, and then image processing is performed; that is, the image is preprocessed, the image is cropped into multiple small images, and data enhancement is performed on the multiple small images; image preprocessing can improve the performance of the model;

[0044] Step S02: training the neural network model and introducing an attention mechanism to enable the network to better understand the relationship between different channels and adaptively adjust the weight of each channel; the neural network model is a residual network, including a convolutional layer, a pooling layer, a residual block structure, a fully connected layer, and an output layer;

[0045] The residual network is a deep convolutional neural network that introduces a skip connection mechanism to form residual connections, achieving more efficient information transfer and a deeper network structure. The residual network consists of two parts: residual blocks and skip connections. The residual block is the basic building block of the residual network. Its core idea is to introduce skip connections to directly add the input data to the output data of the convolutional layer to achieve more efficient information transfer. At the same time, the attention mechanism is introduced to learn the weights of each channel, thereby ensuring the importance of the network in the channel, increasing the weights of channels with effective characteristics, and suppressing ineffective channels.

[0046] After convolution processing, the convolution layer can obtain a feature image. Multiple feature maps constitute a convolution layer. All neurons in each feature map share the same convolution kernel parameters to improve the robustness of the neural network. The feature extraction of the bottom layer is affected by the observation range of the top feature map. The feature maps obtained by different convolution algorithms are also different. In the same area, due to the difference in convolution kernels, the eigenvalues ​​obtained are also different.

[0047] If the convolution kernel k convolves the input data to obtain a feature map, the formula is expressed as: ,in, is the j-th feature map of the l-th layer; is the output value of the previous layer, i.e. the l-1 layer; is the bias value; M j is the area where the convolution kernel acts on the image; Represents a convolution kernel; f represents a nonlinear activation function;

[0048] The size of the output feature map of each layer of convolution depends on the padding value, size, stride, and the size of the input image of the layer. The size of the output feature map of the convolution is expressed as: , where S output is the output feature map size of the convolution, S input is the size of the input image, F is the size of the convolution kernel, P is the number of padded pixels, and B is the sliding step size of the convolution kernel;

[0049] The pooling layer operation consists of a fixed window based on the stride size. The window traverses each position and calculates the output. The purpose of the pooling layer is to reduce the dimensionality of the data, remove some irrelevant information, reduce the number of parameters, and improve computational efficiency. Pooling can not only maintain the local invariance of the image, but also improve the fault tolerance of the image and avoid excessive fitting.

[0050] The formula of the pooling layer is expressed as: ,in, is the input feature of the lth layer, and the output feature after pooling is obtained through pooling operation ; The pooling operation includes but is not limited to maximum pooling and average pooling; the pooling operation requires the pooling core to move along the length and width on the feature map of each channel, and the feature map S output The size of is determined by the hyperparameters of the pooling operation, including but not limited to the pooling window size and stride; the size of the feature map is expressed as follows: , where D output D is the size of the output feature map of the pooling layer. input is the length or width of the input feature map, P is the pixel to be filled, and D filter is the length or width of the pooling kernel, B filter is the moving step of the pooling kernel;

[0051] The formula of the residual block structure is: H(x)=F(x)+x, where x is the input data, F(x) is the residual function, which is the actual learning part of the model; H(x) is the output result;

[0052] When F(x) = 0, H(x) = x, which is the so-called identity mapping. The goal of the training process is to make the residual result as close to 0 as possible, so that the accuracy can be maintained as the network depth increases;

[0053] The fully connected layer combines the features extracted by the pooling layer and the convolutional layer for classification; the function expression of the fully connected layer is: , where z i is the output value of the i-th node, k is the number of corresponding output nodes, that is, the total number of corresponding classification types;

[0054] The attention mechanism is the SE attention mechanism, which performs a pooling operation on the input feature map and then inputs it into the fully connected layer. It is input into the next fully connected layer learning using the ReLU function and then outputted through the Sigmoid function, denoted as S'. The formula is expressed as: , where δ is the activation function ReLU, σ is the activation function Sigmoid, W1 and W2 are the weight matrices of the two fully connected layers respectively, , where Y is the number of hidden nodes in the middle layer, c is the number of image channels, , where W and H are the length and width of the image. Each layer of the input feature map undergoes a 2D spatial kernel convolution, and finally c output feature images are obtained to form the feature map u c ;

[0055] Step S03: Based on the trained neural network, the junk image in step S01 is identified and the identification result is output.

[0056] In this embodiment, it should be specifically explained that the classification analysis module calculates the accuracy rate of each type of garbage placement using the following formula:

[0057] , where β tt is the accuracy rate of garbage placement in the tth category, φ tt is the amount of garbage of the tth category put into the tth trash bin, γ t is the total amount of garbage put into the tth garbage bin; if there are T garbage categories, corresponding to T garbage bins, then t=1, 2, 3...T;

[0058] Calculate the percentage of incorrectly placed garbage in each category of trash bins using the following formula: , where Z qt is the placement rate of the qth garbage category in the garbage bin corresponding to the tth garbage category, φ qt is the amount of garbage of the qth category put into the tth garbage bin, γ t The amount of all garbage put into the t-th garbage bin, and q≠t, q=1, 2, 3...T;

[0059] By analyzing the accuracy rate of each category of garbage disposal and the proportion of incorrectly disposed garbage categories, we can obtain the blind spots in people's knowledge about garbage disposal; if the proportion of a certain category of garbage disposed in another category of trash can is too high, people may not be able to distinguish between the two categories of garbage, so it is necessary to popularize garbage classification knowledge among people, to solve the problem of incorrect garbage classification to a certain extent; if the accuracy rate of a certain category of garbage disposal is too low, it means that people do not know enough about this category of garbage, and they need to popularize knowledge about the content of this category of garbage.

[0060] In this embodiment, it should be specifically explained that the calculation formula of the garbage placement accuracy rate of the input personnel by the supervision and evaluation module is: , where λ d is the garbage delivery accuracy of the dth person, µ d is the correct number of garbage disposals by the dth person, E dThe total number of times the dth person puts out garbage; d = 1, 2, 3...N;

[0061] The calculation formula for the overall garbage placement accuracy is: , where λ´ is the overall garbage placement accuracy;

[0062] If λ d <YU1, indicating that the personnel have an incomplete understanding of the categories of waste classification or do not classify their waste. Therefore, it is necessary to persuade the personnel to classify waste and promote knowledge. YU1 is the set personnel supervision threshold. If YU1>50%, the specific value of YU1 can be set by those skilled in the art according to actual conditions. This embodiment does not impose a specific limit on this value.

[0063] If λ´<YU2, it means that the overall awareness of garbage classification in the region is too low, and it is necessary to carry out garbage classification persuasion work or garbage classification knowledge propaganda in the region as a whole to improve the overall awareness and relevant knowledge of garbage classification among the people in the region; YU2 is the set overall supervision threshold, and if YU2>50% is satisfied, the specific value of YU2 can be set by people in this field according to actual conditions, and this embodiment does not make any specific limitation on this specific value.

[0064] In this embodiment, it should be specifically explained that while the human-computer interaction module displays the data through human-computer interaction, it also displays the images of incorrectly placed garbage in the classification analysis module. The displayed content includes the images of incorrectly placed garbage, the incorrect garbage category, and the correct garbage category. This helps the staff to quickly find the corresponding incorrectly placed garbage when processing garbage classification, place it in the correct garbage category, and carry out subsequent garbage disposal work.

[0065] In this embodiment, it should be specifically explained that the data comparison between the neural network model that introduces the attention mechanism for identifying garbage in this embodiment and the existing network model for identifying garbage is shown in the following table:

[0066]

[0067] It can be seen from the experimental results data in the table that the neural network model using the attention mechanism used in this embodiment has an accuracy rate of up to 93% compared with other network models, and the precision is significantly improved; the required number of parameters, calculation amount, and operation time are all less; although the calculation amount and operation time of the AlexNet model are less than those of the network model of this embodiment, its accuracy is low; the time in the table represents the operation time required for the network model.

[0068] In this embodiment, it should be specifically explained that the main difference between this embodiment and the prior art is that this embodiment has a garbage identification module, a classification analysis module and a supervision and evaluation module, which is conducive to identifying the input garbage based on the neural network model, and introduces an attention mechanism in the neural network module, thereby ensuring the importance of the network in the channel, increasing the weight of the channel with effective characteristics, and suppressing the ineffective channel, so as to obtain accurate garbage identification results and assist the garbage disposal personnel to put the garbage into the correct garbage bin; at the same time, a comprehensive analysis and supervision of the garbage classification of the personnel is carried out to obtain the accuracy of each category of garbage and the proportion of incorrectly placed garbage categories, so as to obtain the personnel's garbage classification awareness and the popularization effect of garbage classification knowledge, thereby providing a guiding direction for subsequent supervision work; and according to the values ​​of the input personnel's garbage disposal accuracy rate and the overall garbage disposal accuracy rate, corresponding supervision operations are performed to enhance the personnel's garbage classification awareness and the popularity of garbage classification, and fundamentally improve the garbage classification effect.

[0069] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

[0070] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. An intelligent supervision system for garbage classification and placement, characterized by: It includes information input module, information verification module, garbage identification module, identification result judgment module, voice prompt module, classification situation analysis module, supervision and evaluation module and human-computer interaction module; The information input module is used to input personnel information and store it in the database, and to label the input personnel, which are sequentially recorded as U1, U2, U3...UN, where N is the total number of input personnel; The information verification module is used to verify the information of the delivery personnel, and verify the information of the delivery personnel through face recognition technology, and transmit the personnel information and the number of the input personnel to the supervision and evaluation module; The garbage recognition module recognizes the input garbage based on the neural network model that introduces the attention mechanism, and outputs the recognition result to the recognition result judgment module; The recognition result discrimination module is used to classify and discriminate the recognition result, that is, to compare and discriminate the garbage category corresponding to the recognition result with the category of the garbage bin, and transmit the discrimination result to the voice prompt module; the discrimination result includes two types: "correct classification" and "wrong classification". If the discrimination result is "wrong classification", the image of the garbage bin is transmitted to the classification situation analysis module at the same time; The classification situation analysis module is used to receive data from the recognition result judgment module, store the images of garbage that are put into the wrong place, and count the types of garbage that are put into the wrong place, so as to obtain the accuracy rate of each type of garbage. The voice prompt module is used to receive data from the recognition result determination module and make corresponding voice announcements according to the recognition results; The supervision and evaluation module is used to supervise the personnel's correction of garbage placement. If the personnel does not change the wrong garbage placement category, the personnel information is recorded, the garbage placement accuracy rate of the input personnel and the overall garbage placement accuracy rate are counted, and corresponding supervision operations are performed; The human-computer interaction module is used to perform human-computer interaction display on the data; The garbage identification module identifies the input garbage based on the neural network model, including the following steps: Step S01: Take a picture of the garbage dumped and mark it, which is recorded as U corresponding to the number of the person who dumped it. da , where d = 1, 2, 3...N, a = 1, 2, 3...E, E is the total number of garbage disposals, U da The garbage image is placed by the dth person for the ath time, and then the image is processed; Step S02: training the neural network model and introducing an attention mechanism; the neural network model is a residual network, including a convolutional layer, a pooling layer, a residual block structure, a fully connected layer and an output layer; Step S03: Based on the trained residual network, identify the garbage image in step S01 and output the identification result; After convolution processing, the convolution layer can obtain a feature image. Multiple feature maps form a convolution layer. All neurons in each feature map share the same convolution kernel parameters. The feature maps obtained by using different convolution algorithms are also different. If the convolution kernel k convolves the input data to obtain a feature map, the formula is expressed as: ,in, is the jth feature map of the lth layer; is the output value of the previous layer, i.e. the l-1th layer; is the bias value; M j is the area where the convolution kernel acts on the image; represents a convolution kernel; f represents a nonlinear activation function; The size of the output feature map of each layer of convolution depends on the padding value, size, stride, and the size of the input image of the layer. The size of the output feature map of the convolution is expressed as: , where S output is the output feature map size of the convolution, S input is the size of the input image, F is the size of the convolution kernel, P is the number of padded pixels, and B is the sliding step size of the convolution kernel; The formula of the pooling layer is expressed as: ,in, is the input feature of the lth layer, and the output feature after pooling is obtained ; The pooling operation requires the pooling kernel to move along the length and width on the feature map of each channel, and the feature map D output The size of is determined by the hyperparameters of the pooling operation, and the size of the feature map is expressed as: , where D output is the size of the feature map output by the pooling layer, D input is the length or width of the input feature map, P is the pixel to be filled, and D filter is the length or width of the pooling kernel, B filter is the moving step of the pooling kernel; The formula of the residual block structure is: H(x)=F(x)+x, where x is the input data, F(x) is the residual function, which is the real learning part of the model; H(x) is the output result. When F(x)=0, H(x)=x, which is the so-called identity mapping. The goal of the training process is to make the residual result as close to 0 as possible. The fully connected layer combines the features extracted by the pooling layer and the convolutional layer for classification; the function expression of the fully connected layer is: , where z i is the output value of the i-th node, k is the number of corresponding output nodes, that is, the total number of corresponding classification types; The attention mechanism is the SE attention mechanism, which performs a pooling operation on the input feature map, and then inputs it into the fully connected layer, uses the ReLU function to input it into the next fully connected layer learning, and then outputs it through the Sigmoid function, which is recorded as S'; the formula is expressed as: , where δ is the activation function ReLU, σ is the activation function Sigmoid, W1 and W2 are the weight matrices of the two fully connected layers respectively, , where Y is the number of hidden nodes in the middle layer, c is the number of image channels, , where W and H are the length and width of the image. Each layer of the input feature map undergoes a convolution with a 2D spatial kernel, and finally c output feature images are obtained to form a feature map u c ; The classification analysis module calculates the accuracy of each type of garbage placement using the following formula: , where β tt is the accuracy of garbage placement in the tth category, φ tt is the amount of garbage of the tth category put into the tth trash can, γ t is the amount of all garbage put into the tth garbage bin; if there are T garbage categories, corresponding to T garbage bins, then t=1, 2, 3...T; Calculate the percentage of incorrectly placed garbage in each category of garbage bins using the following formula: , where Z qt is the placement rate of the qth garbage category in the garbage bin corresponding to the tth garbage category, φ qt is the amount of garbage of the qth category put into the tth garbage bin, γ t The amount of all garbage put into the tth garbage bin, and q≠t, q=1, 2, 3...T.

2. The intelligent supervision system for garbage classification and delivery according to claim 1 is characterized in that: The calculation formula of the supervision and evaluation module for the garbage placement accuracy of the input personnel is: , where λ d is the garbage delivery accuracy of the dth person, µ d is the correct number of garbage disposals by the dth person, E d The total number of times the dth person puts out garbage; d = 1, 2, 3...N; The calculation formula for the overall garbage placement accuracy is: , where λ´ is the overall garbage placement accuracy.