Garbage automatic identification method and system for improving working efficiency of automatic sweeper

Abstract

The application discloses a garbage automatic recognition method and system for improving the working efficiency of an automatic sweeper, and the method comprises the following steps: obtaining a garbage sample library that has been subjected to data labeling; deploying a completed garbage recognition model into an edge algorithm box to perform real-time garbage recognition and result reporting; receiving new labeled garbage sample data to obtain a new garbage recognition model; and evaluating the obtained model through a prediction result when the new garbage recognition model is obtained each time, updating the model with the highest model evaluation index in the edge algorithm box to complete automatic updating of the model. The application enables an unmanned sweeper to realize real-time sensing of the surrounding environment, intelligently recognize the garbage condition around the sweeper, and thus realize more efficient and intelligent cleaning work. Meanwhile, the application can not only realize accurate recognition of garbage, but also can be self-updated to realize recognition of more garbage types.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned sweeping machines, in particular to a garbage automatic identification method and system for improving the working efficiency of automatic sweeping machines and related products. BACKGROUND

[0002] In recent years, with the rapid development of China's economy, the steady improvement of urbanization and urbanization, and the continuous improvement of urban infrastructure, people's demand for sanitation is also increasing. The city population is constantly concentrated, and the city garbage is also increasing, making the cleaning work increasingly heavy, and people have put forward higher standards for cleaning work. Therefore, traditional manual cleaning gradually cannot meet these needs, and unmanned automatic sweeping machines are increasingly needed in the current urban environmental protection work.

[0003] Unmanned automatic sweeping machines not only can clean the road garbage, but also can greatly reduce the labor cost. At the same time, the state has introduced relevant policies to encourage urban environmental protection mechanization. Therefore, unmanned automatic sweeping machines will become the most important cleaning tool in urban environmental protection work.

[0004] However, most of the current sweeping machines work in a fixed mode, which is low in working efficiency. Whether there is garbage on the road or not, the sweeping machine will start the cleaning system to clean the road, resulting in low working efficiency and energy waste.

[0005] Therefore, it is necessary to provide a garbage automatic identification method and system for improving the working efficiency of automatic sweeping machines and related products to overcome the above-mentioned defects. SUMMARY

[0006] The purpose of the present application is to provide a garbage automatic identification method and system for improving the working efficiency of automatic sweeping machines and related products, aiming to solve the problem of low working efficiency of existing unmanned sweeping machines and improve energy utilization.

[0007] In order to achieve the above-mentioned purpose, the first aspect of the present application provides a garbage automatic identification method for improving the working efficiency of automatic sweeping machines, comprising the following steps:

[0008] Step S10: obtaining a garbage sample library that has been labeled with data;

[0009] Step S20: building a garbage identification model based on a deep learning neural network according to garbage identification requirements, and inputting the garbage sample library into the garbage identification model for a preset number of training rounds;

[0010] Step S30: deploying the trained garbage identification model to an edge algorithm box, and then installing the edge algorithm box into an unmanned sweeping machine for real-time garbage identification and result reporting;

[0011] Step S40: receiving new labeled sample data of garbage, and training the garbage identification model according to the new garbage sample data for a preset number of times to obtain a new garbage identification model;

[0012] Step S50: when a new garbage identification model is obtained each time, the obtained model is evaluated through a prediction result, and the model with the highest model evaluation index is updated in the edge algorithm box to complete automatic updating of the model.

[0013] In a preferred embodiment, in step S40, the model updating time is divided into three stages in advance, wherein,

[0014] The first stage includes:

[0015] The garbage sample data collected and labeled by humans is obtained, and then the results of manual model parameter adjustment and network fine-tuning are received, and model prediction and garbage identification are performed;

[0016] The second stage includes:

[0017] A semi-automatic labeling module is embedded in the edge algorithm box to semi-automatically label the newly added categories in the garbage sample data;

[0018] Model training is performed according to the newly added garbage sample data, and then the results of automatic parameter adjustment and structure adjustment realized through the network are received, and the model is overall evaluated and regularly updated;

[0019] The third stage includes:

[0020] The garbage data is automatically collected by the camera of the unmanned sweeper and uploaded to the server;

[0021] The pre-built automatic labeling module is obtained to automatically label the uploaded garbage data, and the server constructs a garbage sample database;

[0022] The server regularly performs model training and self-updating according to the garbage sample database to realize the autonomous learning function of the garbage identification model.

[0023] In a preferred embodiment, it further includes the following steps:

[0024] The inspection results of the garbage sample database and the model identification effect by the staff are received.

[0025] In a preferred embodiment, the implementation steps of the automatic labeling module are:

[0026] A data import module is built to realize automatic loading of data;

[0027] Building a deep learning neural network;

[0028] Loading a pre-trained model;

[0029] Using the pre-trained model to predict the to-be-labeled garbage picture;

[0030] Output and save the prediction results including the picture and the label.

[0031] In a preferred embodiment, in step S10, the garbage sample library is obtained by taking pictures of garbage through a pre-installed camera, and then processing the taken video and labeling garbage data.

[0032] In a preferred embodiment, the step S50 comprises:

[0033] The current garbage recognition model is predicted according to a preset verification set, and the corresponding current garbage recognition model is evaluated according to a preset evaluation formula according to the prediction result, to obtain a model evaluation index ; wherein,

[0034] , is a preset weight, and F is a balance value; wherein,

[0035] , , , , AP is the accuracy of predicting a certain category, m is the correct number of predicting a certain category, n is the total number of labels of a certain category in the training data, mAP is the average accuracy of predicting all categories, N is the number of categories, p is the precision, TP is the correct number of predicting positive samples, FP is the incorrect number of predicting positive samples, R is the recall, TP is the correct number of predicting positive samples, and FN is the incorrect number of predicting negative samples.

[0036] In a preferred embodiment, the data types labeled in the garbage sample library include one or a combination of the following: fallen leaves, cigarette butts, plastic bottles, waste paper, paper boxes, water pipes, masks, electric wires, sewage, petals, ditches, pits, dog feces, and films.

[0037] The second aspect of the application is an automatic garbage recognition system for improving the working efficiency of an automatic sweeper, comprising:

[0038] A data labeling module is configured to obtain a garbage sample library that has been labeled.

[0039] A model building module is configured to build a garbage recognition model based on a deep learning neural network according to garbage recognition requirements, and input the garbage sample library into the garbage recognition model for a preset number of training rounds.

[0040] a model installation module, configured to deploy the trained garbage recognition model to an edge algorithm box, and then install the edge algorithm box to the unmanned sweeper, so as to perform real-time garbage recognition and result reporting;

[0041] a model updating module, configured to receive new labeled garbage sample data, and perform preset rounds of training on the garbage recognition model according to the new garbage sample data, so as to obtain a new garbage recognition model;

[0042] a model evaluation module, configured to evaluate the obtained model through a prediction result when a new garbage recognition model is obtained each time, and update the model with the highest model evaluation index in the edge algorithm box, so as to complete automatic updating of the model.

[0043] The third aspect of the present application provides a terminal, which comprises a memory, a processor and a computer program stored in the memory, and the computer program is executed by the processor to implement each step of the garbage automatic recognition method for improving the working efficiency of the automatic sweeper according to any one of the above embodiments.

[0044] The fourth aspect of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement each step of the garbage automatic recognition method for improving the working efficiency of the automatic sweeper according to any one of the above embodiments.

[0045] The fifth aspect of the present application provides a computer program product, which comprises a computer program or instructions, and the computer program or instructions are executed by a processor to implement each step of the garbage automatic recognition method for improving the working efficiency of the automatic sweeper according to any one of the above embodiments.

[0046] The garbage automatic recognition method, system and related products for improving the working efficiency of the automatic sweeper provided by the present application enable the unmanned sweeper to perceive the surrounding environment in real time and intelligently recognize the garbage condition around. When the sweeper recognizes the garbage, it actively starts the cleaning work, and when there is no garbage, it stops the cleaning work, so as to realize more efficient and intelligent cleaning work. At the same time, the present application not only enables accurate recognition of garbage, but also enables self-updating, realizes recognition of more types of garbage, so that the garbage recognition model has better robustness and generalization ability, and realizes truly intelligent cleaning. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. It should be understood that the following drawings only show some of the embodiments of the present application, and therefore should not be regarded as a limitation on the scope. Other related drawings can also be obtained by those of ordinary skill in the art without creative labor on the basis of these drawings.

[0048] Figure 1 is a convolution schematic diagram;

[0049] Figure 2 is a maximum pooling schematic diagram;

[0050] Figure 3 is a residual structure schematic diagram;

[0051] Figure 4 is a flowchart of the garbage automatic identification method for improving the working efficiency of the automatic sweeper provided by the present application;

[0052] Figure 5 is a framework diagram of the garbage automatic identification system for improving the working efficiency of the automatic sweeper provided by the present application. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and beneficial technical effects of the present application more clear and understandable, the present application will be further described in detail below in combination with the drawings and specific embodiments. It should be understood that the specific embodiments described in the present specification are only for the purpose of explaining the present application, and are not intended to limit the present application.

[0054] It should also be understood that the terms used in the present application specification herein are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the present application specification and the appended claims, the singular forms "a", "an" and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0055] It should also be further understood that the term "and / or" used in the present application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0056] Embodiment one

[0057] In the embodiments of the present application, a garbage automatic identification method for improving the working efficiency of an automatic sweeper is provided, which is used to help the unmanned sweeper to perceive the surrounding environment in real time and intelligently identify the garbage condition around. When the sweeper identifies garbage, it actively starts the cleaning work, and when there is no garbage, it closes the cleaning work.

[0058] Before introducing the technical solutions of the present application, the following introduction is made to the algorithms related to the deep learning network building:

[0059] (1) Convolution layer: the role of the convolution layer is to extract the features of the input data by performing convolution operation on the input data with a convolution kernel in a sliding window manner. The convolution operation is shown in FIG. 1, wherein each weight value in the convolution kernel is multiplied by the corresponding element value in the sliding window in the input data, and then the multiplication result is accumulated, and then the calculation result is saved to the corresponding output position, and the process is repeated with a specified step until the input data is traversed. Figure 1

[0060] (2) Pooling layer: the edge features extracted by the convolution layer are too accurate, and too much information is retained, but sometimes the actual feature extraction task does not want the input information to be retained too much. The pooling layer can count the adjacent regions of the feature map to achieve the aggregation effect, which is actually a sampling process. After the pooling operation, the sensitivity of the convolution layer to the edge features is reduced, and the data is also reduced in dimension, achieving the effects of reducing the number of parameters in the network and reducing the calculation amount of the entire network. In addition, overfitting can also be effectively reduced. Therefore, the pooling operation is usually placed after the convolution operation to further strengthen the feature extraction capability. Pooling can be divided into maximum pooling and average pooling. Figure 2 The maximum pooling operation calculation process is shown in FIG. 2. The mathematical formula of the pooling feature extraction is as follows:

[0061] (4-1)

[0062] wherein, is the value of the spatial coordinate [i, j] after convolution, f[m, n] is the value of the original picture space coordinate [m, n], and g[i-m, i-n] is the value of the convolution kernel space coordinate [i-m, i-n];

[0063] (3) Residual structure: a series of residual structures form a residual network. The residual structure has two operation paths, i.e., a main branch and a side branch. The residual structure is shown in FIG. 3. Figure 3 wherein x is the output feature map of the previous layer and is used as the input of the residual structure, and the output of the residual structure can be expressed by formula (4-1):

[0064] y = F(x, w) + x (4-2)

[0065] wherein F(x, w) is the output of the residual part to x, the side branch x and F(x, w) are combined in the channel layer, and the final output can be expressed as:

[0066] y = F(x, w) + h(x) (4-3)

[0067] ​Residual structure is widely used in residual network, and the final can solve the problem of gradient disappearance to some extent.

[0068] In the embodiments of the present application, as shown in Figure 4 The garbage automatic recognition method for improving the working efficiency of the automatic sweeper includes steps S10-S50.

[0069] Step S10: Obtain a garbage sample library that has been data-labeled.

[0070] Specifically, the garbage sample library is obtained by taking pictures of garbage through a pre-installed camera, and then processing the video and labeling garbage data. The data types labeled in the garbage sample library include one or a combination of the following: fallen leaves, cigarette butts, plastic bottles, waste paper, paper boxes, water pipes, masks, electric wires, sewage, petals, ditches, pits, dog feces, film, etc.

[0071] For example, before taking pictures of garbage, the camera needs to be selected according to the required distance, light resistance, waterproofness, field of view, etc. For example, a 500W pixel, 6mm focal length, wide dynamic 85, USB interface, high-definition roller shutter door camera can be selected. Then determine the installation position of the camera on the unmanned sweeper, for example, it can be deployed in front of the vehicle head of the unmanned sweeper, and set towards the ground at a preset angle, so as to capture the picture within a certain range in front of the vehicle head.

[0072] Step S20: According to the garbage recognition requirement, a garbage recognition model based on a deep learning neural network is built, and the garbage sample library is input into the garbage recognition model for preset rounds of training.

[0073] Specifically, first, a data loading module is developed to realize the automatic loading function of data; second, a 64-layer deep convolutional neural network is built using convolution, activation function and other operators; finally, the loaded data based on the labeled garbage sample library is divided into a training set and a validation set, the training set data is input into the neural network, and the network is trained based on random gradient descent, maximum likelihood estimation and other algorithms to obtain the garbage recognition model.

[0074] Step S30: Deploy the trained garbage recognition model into an edge algorithm box, and then install the edge algorithm box into the unmanned sweeper for real-time garbage recognition and result reporting.

[0075] For example, the obtained garbage recognition model is deployed into an edge algorithm box (such as an RK3588s algorithm box) for real-time garbage recognition and result reporting. The industrial computer of the unmanned sweeper automatically generates a cleaning strategy according to the perceived surrounding garbage, and performs cleaning work, that is, initiates cleaning work when garbage is recognized, and closes the cleaning work when there is no garbage.

[0076] Step S40: receiving new labeled garbage sample data, and training the garbage recognition model according to the newly added garbage sample data for a preset number of times to obtain a new garbage recognition model.

[0077] Since the types and shapes of the garbage pictures obtained by the panning shots cannot cover all garbage situations in reality, the unmanned sweeping robot will also input the photographed to-be-recognized images as a new training set into the garbage recognition model for training when performing the sweeping work, so as to form a benign positive cycle to continuously improve the prediction accuracy and generality of the garbage recognition model.

[0078] In step S40, the model updating time is divided into three stages, i.e., an early stage, a middle stage and a late stage.

[0079] The first stage includes: obtaining garbage sample data collected by humans and labeled by humans, then receiving the results of the model parameter adjustment and network fine-tuning performed by humans, and performing model prediction and garbage recognition. That is, in the early stage, the garbage pictures in the sample set (i.e., the training set) for training are labeled by humans, and the model parameters are adjusted by humans, and the network is fine-tuned, so that the initial garbage recognition model has certain garbage recognition capability.

[0080] The second stage includes: embedding a semi-automatic labeling module in the edge algorithm box to semi-automatically label the newly added categories in the garbage sample data; training the model according to the newly added garbage sample data, then receiving the results of the automatic parameter adjustment and structure adjustment realized by the network, and performing overall evaluation and regular model updating on the model. That is, at this time, the unmanned sweeping robot will photograph a large number of actual scene images during the cleaning operation, which may include some new garbage types. At this time, the images that can be predicted are automatically labeled, and the images that the model cannot recognize are arranged to be regularly checked by personnel and supplemented by human data labeling.

[0081] The third stage includes: the unmanned sweeping robot automatically collects garbage data through its own camera and uploads it to the server; obtaining a pre-built automatic labeling module to automatically label the uploaded garbage data, and the server constructs a garbage sample database; the server regularly trains and updates the model according to the garbage sample database to realize the self-learning function of the garbage recognition model. That is, after a certain period of time, the pictures are automatically photographed by the unmanned sweeping robot and uploaded, then automatically labeled by the background server and a continuously updated garbage sample database is constructed, and the server uses the garbage sample database as a new training set to regularly update the garbage recognition model, thereby realizing the self-learning function of the garbage recognition model.

[0082] Further, in order to improve the accuracy of the database, the receiving staff regularly checks the results of the garbage sample database and model recognition effect, thereby eliminating the prediction pictures and labeled data with larger errors.

[0083] The implementation steps of the automatic labeling module are as follows:

[0084] (1) Build a data import module to realize automatic loading of data;

[0085] (2) Build a deep learning neural network;

[0086] (3) Load a pre-trained model;

[0087] (4) Use the pre-trained model to predict the garbage pictures to be labeled;

[0088] (5) Output and save the prediction results including the picture image and the label label.

[0089] Of course, the implementation principle of the automatic labeling module for automatically labeling images can also refer to the prior art, which will not be repeated here.

[0090] Step S50: When a new garbage recognition model is obtained, the obtained model is evaluated through the prediction results, the model with the highest model evaluation index is updated in the edge algorithm box, so as to complete the automatic updating of the model.

[0091] Therefore, the application constructs a model evaluation system, and after the first garbage recognition model is obtained and the model is updated each time, the prediction results of the model are evaluated through the prediction set, a quantifiable model evaluation index is obtained, and then it is judged whether the garbage recognition model in the algorithm box is updated based on the model evaluation index. If the index improves, the model is updated; if the index decreases, the current model is maintained.

[0092] Specifically, step S50 includes:

[0093] The current garbage recognition model is predicted according to the preset verification set, and the corresponding current garbage recognition model is evaluated according to the preset evaluation formula based on the prediction result, so as to obtain a model evaluation index ; wherein,

[0094] , is a preset weight, and F is a balance value; wherein,

[0095] , , , , , AP is the accuracy of predicting a certain category, m is the correct number of predicting a certain category, n is the total number of labels of a certain category in the training data, mAP is the average accuracy of predicting all categories, N is the number of categories, p is the precision, TP is the correct number of predicted positive samples, FP is the wrong number of predicted positive samples, R is the recall, TP is the correct number of predicted positive samples, and FN is the wrong number of predicted negative samples.

[0096] Therefore, the present application can not only realize the accurate identification of garbage by the unmanned sweeper, but also can be self-updated, realize the identification of more garbage types, so that the garbage identification model has better robustness and generalization ability, and realizes the truly intelligent cleaning.

[0097] Example two

[0098] The garbage automatic identification system 100 for improving the working efficiency of the automatic sweeper is used to help the unmanned sweeper to perceive the surrounding environment in real time and intelligently identify the garbage condition around. It should be noted that the implementation principle and specific implementation manner of the garbage automatic identification system 100 for improving the working efficiency of the automatic sweeper can refer to the garbage automatic identification method for improving the working efficiency of the automatic sweeper described above, and therefore will not be described here.

[0099] As shown in Figure 5 The garbage automatic identification system 100 for improving the working efficiency of the automatic sweeper comprises:

[0100] The data labeling module 10 is used to obtain a garbage sample library that has been labeled with data;

[0101] The model building module 20 is used to build a garbage identification model based on a deep learning neural network according to garbage identification requirements, and input the garbage sample library into the garbage identification model for training for a preset number of rounds;

[0102] The model installation module 30 is used to deploy the trained garbage identification model to an edge algorithm box, and then install the edge algorithm box into the unmanned sweeper to perform real-time garbage identification and result reporting;

[0103] The model updating module 40 is used to receive new labeled garbage sample data, and train the garbage identification model for a preset number of rounds according to the new garbage sample data to obtain a new garbage identification model;

[0104] The model evaluation module 50 is used to evaluate the obtained model through the prediction result when a new garbage identification model is obtained, update the model with the highest model evaluation index in the edge algorithm box to complete the automatic updating of the model.

[0105] Example three

[0106] The application provides a terminal, which comprises a memory, a processor and a computer program stored in the memory, and the computer program realizes each step of the garbage automatic identification method for improving the working efficiency of an automatic sweeper when executed by the processor.

[0107] Embodiment four

[0108] The application provides a computer readable storage medium, which stores a computer program, and the computer program realizes each step of the garbage automatic identification method for improving the working efficiency of an automatic sweeper when executed by a processor.

[0109] Embodiment five

[0110] The application provides a computer program product, which comprises a computer program or instructions, and the computer program or instructions realize each step of the garbage automatic identification method for improving the working efficiency of an automatic sweeper when executed by a processor.

[0111] To sum up, the garbage automatic identification method for improving the working efficiency of an automatic sweeper, the system and the related product provided by the application enable the unmanned sweeper to perceive the surrounding environment in real time and intelligently identify the garbage condition around. When the sweeper identifies the garbage, the sweeper initiates the cleaning work, and when there is no garbage, the sweeper stops the cleaning work, so that more efficient and intelligent cleaning work is realized. Meanwhile, the application can not only realize accurate garbage identification, but also can be updated automatically, so that more garbage types are identified, so that the garbage identification model has better robustness and generalization ability, and truly intelligent cleaning is realized.

[0112] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional units and modules is exemplified, and in actual application, the above functions can be completed by different functional units and modules according to needs, that is, the internal structure of the system is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit or module in the embodiment can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit, and the integrated unit can be realized in the form of hardware or software. In addition, the specific names of each functional unit or module are only for convenient distinction, and do not limit the protection scope of the application. The specific working process of the units or modules in the system can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0113] In the above embodiments, the description of each embodiment is focused on, and the part not described or recorded in a certain embodiment can be referred to the relevant description of other embodiments.

[0114] Those skilled in the art can realize that the units and method steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0115] In the embodiments provided by the present application, it should be understood that the disclosed system or device / terminal equipment and method can be implemented by other ways. For example, the system or device / terminal equipment embodiments described above are only schematic, and the division of the modules or units is only a logical function division, and there can be another division manner in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between the units can be indirect couplings or communication connections through some interfaces, and can be electrical, mechanical or other forms.

[0116] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e. can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

[0117] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or software functional unit.

[0118] The present application is not limited to the description and embodiments described in the specification, and therefore those skilled in the art can easily realize other advantages and modifications, and therefore the present application is not limited to specific details, representative devices and examples of the drawings shown and described herein. The spirit and scope of the general concept defined by the claims and their equivalents are not limited.

Claims

1. A method for automatically identifying garbage to improve the working efficiency of an automatic sweeper, characterized in that, Comprising the following steps: Step S10: Obtain a garbage sample library that has been data-labeled; Step S20: Build a garbage recognition model based on a deep learning neural network according to garbage recognition requirements, and input the garbage sample library into the garbage recognition model for preset rounds of training; Step S30: Deploy the trained garbage recognition model into an edge algorithm box, and then install the edge algorithm box into an unmanned sweeper to perform real-time garbage recognition and result reporting; Step S40: Receive new garbage sample data that has been labeled, and perform preset rounds of training on the garbage recognition model according to the new garbage sample data to obtain a new garbage recognition model; Step S50: When a new garbage recognition model is obtained each time, evaluate the obtained model through a prediction result, update the model with the highest model evaluation index in the edge algorithm box, to complete automatic updating of the model; In step S40, the model updating time is divided into three stages in advance, wherein, The first stage includes: Obtaining garbage sample data that is artificially collected and data-labeled, and then receiving results of model parameter adjustment and network fine-tuning by humans, to perform model prediction and garbage recognition; The second stage includes: Embedding a semi-automatic labeling module in the edge algorithm box to semi-automatically label new categories in the garbage sample data; Performing model training according to the new garbage sample data, and then receiving results of automatic parameter adjustment and structure adjustment through a network, to perform overall evaluation of the model and regular model updating; The third stage includes: The unmanned sweeper automatically collects garbage data through its own camera and uploads the garbage data to a server; Obtaining a pre-built automatic labeling module to automatically label the uploaded garbage data, and the server builds a garbage sample database; The server regularly performs model training and self-updating according to the garbage sample database, to realize a self-learning function of the garbage recognition model; The step S50 includes: The current garbage recognition model predicts according to a preset verification set, and evaluates the corresponding current garbage recognition model according to a prediction result through a preset evaluation formula, to obtain a model evaluation index ; wherein, , is a predetermined weight, F is a balance value; wherein, , , , , , AP is the accuracy of predicting a certain class, m is the correct number of predicting a certain class, n is the total number of labels of a certain class in the training data, mAP is the average accuracy of predicting all classes, N is the number of classes, p is the precision, TP is the correct number of predicting positive samples, FP is the wrong number of predicting positive samples, R is the recall, TP is the correct number of predicting positive samples, FN is the wrong number of predicting negative samples.

2. The garbage automatic recognition method for improving the working efficiency of the automatic sweeper according to claim 1, characterized in that, Further comprising the steps of: Receiving inspection results of the garbage sample database and model recognition effect by a worker on a regular basis.

3. The garbage automatic recognition method for improving the working efficiency of the automatic sweeper according to claim 1, characterized in that, The implementation steps of the automatic labeling module are: Building a data import module to realize automatic loading of data; Building a deep learning neural network; Loading a pre-trained model; Using the pre-trained model to predict garbage pictures to be labeled; Outputting and saving prediction results including pictures and labels.

4. The garbage automatic recognition method for improving the working efficiency of the automatic sweeper according to claim 1, characterized in that, In step S10, the garbage sample library is obtained by taking pictures of garbage through a pre-installed camera, and then processing the taken videos and performing garbage data labeling.

5. The garbage automatic recognition method for improving the working efficiency of the automatic sweeper according to claim 1, characterized in that, The types of data labeled in the garbage sample library include one or a combination of the following: fallen leaves, cigarette butts, plastic bottles, waste paper, paper boxes, water pipes, masks, electric wires, sewage, petals, ditches, pits, dog feces, and films.

6. A garbage automatic recognition system for improving the working efficiency of an automatic sweeper, characterized in that, Comprise: A data labeling module for obtaining a garbage sample library that has been data-labeled; A model building module for building a garbage recognition model based on a deep learning neural network according to garbage recognition requirements, and inputting the garbage sample library into the garbage recognition model for preset rounds of training; A model installation module is configured to deploy the trained garbage recognition model to an edge algorithm box, and then install the edge algorithm box in the unmanned sweeper to perform real-time garbage recognition and result reporting. A model updating module is configured to receive new labeled garbage sample data, and train the garbage recognition model according to the new garbage sample data for a preset number of times to obtain a new garbage recognition model. A model evaluation module is configured to evaluate the obtained model through prediction results when a new garbage recognition model is obtained, update the model with the highest model evaluation index in the edge algorithm box to complete automatic updating of the model. The application of the model updating module specifically includes: The model updating time is divided into three stages, wherein, The first stage includes: Obtaining garbage sample data collected and labeled by humans, receiving results of model parameter adjustment and network fine-tuning by humans, and performing model prediction and garbage recognition. The second stage includes: Embedding a semi-automatic labeling module in the edge algorithm box to semi-automatically label new categories in the garbage sample data; Training the model according to the new garbage sample data, receiving results of automatic parameter adjustment and structure adjustment through the network, and performing overall evaluation of the model and regular model updating. The third stage includes: The unmanned sweeper automatically collects garbage data through its camera and uploads the data to a server; Obtaining a pre-built automatic labeling module to automatically label the uploaded garbage data, and the server builds a garbage sample database; The server regularly trains and updates the model according to the garbage sample database to realize the self-learning function of the garbage recognition model. The application of the model evaluation module specifically includes: The current garbage recognition model predicts according to a preset verification set, and evaluates the corresponding current garbage recognition model according to a prediction result through a preset evaluation formula to obtain a model evaluation index ; wherein, , is a predetermined weight, F is a balance value; wherein, , , , , , AP is the accuracy of predicting a certain class, m is the correct number of predicting a certain class, n is the total number of labels of a certain class in the training data, mAP is the average accuracy of predicting all classes, N is the number of classes, p is the precision, TP is the correct number of predicting positive samples, FP is the wrong number of predicting positive samples, R is the recall, TP is the correct number of predicting positive samples, FN is the wrong number of predicting negative samples.

7. A terminal, characterized by comprising: The terminal includes a memory, a processor, and a computer program stored in the memory, and the computer program is executed by the processor to realize each step of the garbage automatic recognition method for improving the working efficiency of the automatic sweeper according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to realize each step of the garbage automatic recognition method for improving the working efficiency of the automatic sweeper according to any one of claims 1-5.

Images