CN111619992A - Intelligent garbage classification system and method based on machine vision - Google Patents

Abstract

The invention discloses an intelligent garbage classification system and method based on machine vision, which are characterized in that a conveying device is designed and realized by applying the MobileNet deep learning and machine vision technology and adopting Raspberry pi and STM32 as main controllers, so that the full-automatic garbage throwing is realized without manpower input. The test result shows that the key performance of the product is greatly improved compared with that of the classification garbage can on the market, and the method successfully realizes the following steps: (1) the method is suitable for different temperatures and illumination environments and accurate judgment; (2) the recognition precision is improved to more than 95 percent; (3) the method has excellent processing capability, can complete detection of 128 points at most within 7 seconds at high speed, and identifies the speed block.

Description

A machine vision-based intelligent garbage classification system and method

technical field

The invention relates to an intelligent garbage classification system and method based on machine vision.

Background technique

With the development of economy and society, the amount of domestic waste in my country has increased rapidly, and the environmental problems caused by it have become increasingly prominent, which has become a restrictive factor for the development of new urbanization. Garbage classification is an important link and key area in the construction of ecological civilization. my country is forcing the establishment of a domestic waste classification management system, and various provinces and cities have responded one after another. Changsha City will also officially implement waste classification in October this year. Before 2019, 8 cities in my country had carried out pilot projects for waste sorting and collection, but the popularization effect was not satisfactory. The reasons for the ineffective implementation of garbage classification include: (1) the lack of residents' awareness of garbage classification; (2) the inability to link together the collection, transportation, and disposal of garbage after it is put in.

The intelligent classification trash can is designed to realize the automation and intelligence of garbage classification and effectively promote the popularization of garbage classification. However, the few existing system solutions at home and abroad are not very practical and have a single function. The common problems include: (1) the recognition accuracy is not high, resulting in classification errors; (2) the recognition speed is too slow; (3) the temperature is affected The environmental impact such as , light, etc. is obvious, and the performance is unstable; (4) lack of detailed data analysis, and it is difficult to efficiently connect with the recycling process. Among them, the low recognition accuracy and the slow recognition speed are the key pain points that this type of trash can is difficult to use.

Therefore, it is necessary to design an intelligent garbage classification system and method based on machine vision.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to provide an intelligent garbage classification system and method based on machine vision, which can realize automatic detection and classification of garbage.

The technical solution of the invention is as follows:

An intelligent garbage classification system based on machine vision, including a garbage can, and an image recognition garbage classification module, a temporary storage box and a plurality of sorting boxes are arranged at the garbage can;

The image recognition garbage classification module includes MCU, trigger, camera and distribution device; the trigger, camera and distribution device are all connected to the MCU;

The trigger is a pyroelectric infrared sensor, which is used to sense the human body approaching the trash can;

The camera is used to obtain the image of the garbage in the temporary storage box;

The MCU is used to perform image processing on the image and identify the type of garbage;

The sorting device is used to send the garbage from the temporary storage box to the corresponding sorting box according to the sorting instruction issued by the MCU. The MCU has a garbage classification module based on MobileNet and machine vision, which can identify and classify garbage based on the acquired images.

The distribution device includes a conveyor belt (2), a pulley (3) and a conveyor belt driving mechanism for driving the conveyor belt to move;

There are 2 pulleys, the conveyor belt is set on the 2 pulleys, the temporary storage box is fixed on the conveyor belt, the bottom of the temporary storage box is provided with a movable plate and a movable plate driving mechanism for controlling the opening and closing of the movable plate; the sorting box is located below the conveyor belt ;

The sorting device also includes a position detection mechanism for detecting the position of each sorting box;

With the position detection mechanism, when the movable plate is opened, the garbage in the temporary storage box falls into a certain sorting box; there is a hole in the middle of the conveyor belt, so that the garbage from the temporary storage box can enter the sorting box through the hole. .

There are odor sensors and odor deodorizers in the temporary storage box and the sorting box. The odor deodorizing device specifically adopts an ozone generator, which is used to generate ozone to remove odor.

Also includes an overfill reminder module; the overfill reminder module includes an ultrasonic sensor placed in the sorting bin for detecting the height of the waste. When it is detected that the garbage height is greater than the set value, the alarm will be activated.

5. The machine vision-based intelligent garbage classification system according to claim 1, further comprising a server; the MCU in the garbage bin is connected to the server through a communication module, and the operator can access the server through a PC or a mobile phone APP Get spam.

6. An intelligent garbage classification method based on machine vision, characterized in that, adopting the intelligent garbage classification system according to any one of claims 1-5; comprising the following steps:

Step 1: Based on the trigger, detect the garbage thrower approaching the trash can;

Step 2: Start image capture;

After the garbage is put into the temporary storage box, image shooting is started to obtain the garbage image; necessary preprocessing is performed on the garbage image, including grayscale processing, image segmentation processing, etc., which is the prior art.

Step 3: Garbage identification based on garbage images;

Identify garbage based on the Caffe deep learning framework and the neural network MobileNet model;

Step 4: classify the garbage based on the garbage identification result;

Based on the garbage identification results, the identified garbage is transferred from the temporary storage box to the corresponding sorting box.

OpenCV is a cross-platform computer vision library released under the BSD license (open source) and can run on Linux, Windows, Android and Mac OS operating systems. It is lightweight and efficient—consisting of a series of C functions and a small number of C++ classes, and provides interfaces to languages such as Python, Ruby, and MATLAB, and implements many general algorithms in image processing and computer vision.

The present invention proposes the following solutions on the basis of conducting a large number of investigations and analysis for the classification requirements of the current urban domestic waste and the problems faced:

(1) Publish and update information about garbage classification on the WeChat applet in a timely manner; when the intelligent classification trash can completes the identification and classification of the target garbage, it will broadcast and inform users of the type of garbage through voice broadcast, which can popularize garbage classification to citizens and help reduce the cost of garbage classification. Memory and classification difficulty.

(2) Use the big data platform to complete the statistics on the type and quantity of user garbage, and generate a visual report on the web page management end. Through big data analysis, workers can be reminded to clean up, and the best recycling route can be planned for clean vehicles, saving operating costs for sanitation structures.

(3) The use of intelligent garbage sorting devices to achieve accurate identification and separation of dry and wet garbage, high probability to avoid the occurrence of mixed delivery, integrated use of the Internet, big data, Internet of Things and other related technologies, can reasonably plan the incineration cycle of garbage dumps, reduce due to Air pollution caused by incineration.

(4) Install the conveying device for the present invention, after completing the identification and classification of the target garbage, the main control system drives the conveying device, and throws the garbage into the corresponding trash can.

(5) Using the lightweight neural network MobileNet model, the recognition accuracy rate is guaranteed to be above 95%, and the recognition and classification time is controlled within 7 seconds, so as to solve the problems of low accuracy rate, slow recognition speed, and obvious environmental impact. .

Beneficial effects:

The intelligent garbage classification system and method based on machine vision of the present invention uses MobileNet deep learning and machine vision technology, adopts Raspberry pi and STM32 as the main controller, and designs and realizes the transmission device, which realizes the automatic throwing of garbage without manpower input. The test results show that the key performance of the present invention is greatly improved compared with the classified trash cans on the market, and it has successfully achieved: (1) adapting to different temperatures and lighting environments, and making accurate judgments; (2) improving the recognition accuracy to more than 95%; ( 3) With excellent processing ability, it can complete the detection of up to 128 points at high speed within 7 seconds, and the recognition speed is fast.

The WeChat applet of the system is rich in functions, including: (1) uploading unrecognized spam pictures to the background self-training model; (2) one-click query of spam categories; (3) garbage overflow reminder; (4) intelligent message push; (5) Intelligent robot online chat function. The web management terminal uses big data technology to build an intelligent supervision platform, which can realize precise environmental supervision based on data decision-making, management, and service, realize functions such as garbage truck path planning, garbage dump incineration cycle planning, etc. Establishing communication channels for waste sorting and building a household-community waste sorting and recycling network will help to achieve a long-term management mechanism for waste sorting.

The main features and innovations of the present invention are as follows:

(1) Machine vision intelligent garbage classification

In view of the low recognition accuracy and slow recognition speed of the existing "garbage classification artifacts" on the market, this product has developed a machine vision intelligent classification system based on artificial intelligence and big data. Using MobileNet and computer vision, the recognition accuracy rate is as high as more than 95%, and then the recognition return value is returned to the Raspberry Pi main control system.

(2) Intelligent automatic delivery device

At present, the existing smart trash can only remind the user of the garbage category, and the user still needs to manually classify it. The team has developed an intelligent fully autonomous delivery device that communicates with the Raspberry Pi serial port through STM32 to control the forward and reverse rotation of the steering gear. The Raspberry Pi uses the result value returned by the machine vision recognition system to drive the motor and the conveyor belt to put the garbage into the trash can corresponding to the type of garbage.

(3) Odor deodorization

Most trash cans can't keep out odors. Hot weather makes food easily spoiled and fermented, attracting flying insects and possibly emitting toxic gases. In order to solve this practical problem, the intelligent sorting trash can developed by this team is equipped with an odor deodorization device. When the odor and harmful gas are detected by the odor sensor, the sterilization and deodorization module is activated to isolate the odor of garbage and beautify the home environment.

(4) Overfill detection

It often happens that household trash cans are overflowing but users forget to dump. In response to this common phenomenon, the present invention designs a garbage overflow reminder function, detects the garbage height through an ultrasonic sensor, and intelligently reminds the user through a WeChat applet when it overflows, so as to avoid the hazard of garbage hoarding.

(5) Real-time visual monitoring

Considering the weak awareness of residents' garbage classification and the difficulty of community management, the system is based on the Internet of Things technology to effectively perceive and transmit relevant information, and further relies on big data analysis technology on the Web side to count the types of garbage thrown by residents, etc., to achieve garbage classification. Visual monitoring, gradually cultivate users' awareness of garbage classification, and build a family-community management network. Further realization: a. Time route planning of garbage trucks; b. Waste incineration cycle management of garbage incineration plants, and is committed to establishing a refined and efficient garbage recycling system.

Description of drawings

Figure 1 is a functional block diagram;

Fig. 2 is a system topology diagram;

Fig. 3 is a trigger structure diagram;

Fig. 4 is the flow chart of realization scheme;

5 is a structural diagram of a transmission device;

Figure 6 is a diagram of other functional modules;

Figure 7 is a schematic diagram of the conveying mechanism (front view);

FIG. 8 is a schematic diagram (top view) of the conveying mechanism.

Label description: 1-temporary storage box, 2-conveyor belt, 3-belt wheel, 4-sorting box, 11-movable board.

Detailed ways

The present invention will be described in further detail below in conjunction with the accompanying drawings and specific embodiments:

Example 1: System master plan

The system can be divided into five parts:

(1) Image recognition garbage classification;

(2) Garbage disposal device;

(3) Cloud server;

(4) database;

(5) WeChat applet;

(6) Web management background

Realization of Garbage Image Recognition and Classification

Garbage image recognition and classification is the core part of this system. OpenCV is used to process the image to extract the target garbage location and features for subsequent deep learning. The mobile net neural network algorithm is used to classify junk images.

Function composition introduction

Equipment for image recognition garbage classification includes:

(1) Trigger. The main function of the trigger pyroelectric infrared sensor is to obtain information about the proximity of the human body and send a photo request to the Raspberry Pi, thereby triggering the operation of the entire system

(2) Garbage image recognition. Use Raspberry Pi, camera, and OpenCV to complete the image positioning and recognition function.

(3) Garbage classification. The MobileNet model is responsible for garbage classification. If the garbage identification and classification fails, upload the garbage pictures to the database through the WeChat applet.

trigger

The trigger is composed of a pyroelectric infrared sensor, a Raspberry Pi, which is used to sense the proximity signal of the human body. When the distance between the human body and the intelligent sorting trash can reaches the sensing distance threshold, the pyroelectric infrared sensor will detect the change of the infrared temperature of the human body, and transmit a trigger signal to the Raspberry Pi, and then the Raspberry Pi will turn on the camera module to prepare for the opening of the graphic recognition device. .

Pyroelectric infrared sensor: When a person is close to the trash can, the subsequent circuit of the sensor can trigger the switch action after detection and processing, and the sensor will send a signal to the Raspberry Pi at this time.

Raspberry Pi: The development board model used in this system is Raspberry Pi 3B, 1.2GHz, 64-bit processor chip, with low energy consumption and stable operation, which can fully meet the image processing requirements and speed requirements of this system.

Camera module: The input device used in this system is the ov5647 camera without additional driver, which is connected to the Raspberry Pi development board through the CSI interface. This module supports 500w pixel image acquisition, and the target garbage can be located through this module with the code.

Garbage Image Recognition

(1) Target garbage location

Using OpenCV to implement the background difference algorithm, first convert the collected three-channel RGB image into a single-channel grayscale image through the cvtColor function. In order to reduce the error caused by the background illumination in the trash can, the existing grayscale image can be binarized through the threshold function. Then, the binary image containing the target to be recognized is subtracted from the binary background image to realize the background difference.

(2) Target garbage feature extraction

The target garbage to be detected should be clearly distinguishable in the image. In order to ensure the detection accuracy, the MORPH_CLOSE closed operation interface is called through the morphologyEx function to connect the large white area in the background difference result, remove the small area of white noise, and enhance the image features. Finally, the minAreaRect function is used to obtain the minimum enclosing rectangle from the result image of the closing operation. According to the coordinates of the rectangle, the region of interest is set in the original image, and the target to be detected is cut out for subsequent deep learning.

(3) Background difference algorithm

The detection of target garbage by using the background difference algorithm is mainly divided into four steps: background modeling, background update, target garbage detection, and post-processing. Among them, the method of background modeling adopts a single Gaussian distribution model: The gray value of each pixel in the image is regarded as a random process x, x is the gray value, and the gray value of a pixel at this point is assumed. The probability of occurrence obeys a Gaussian distribution and can be expressed as:

δt is the standard deviation of _x , ^μt is the expectation of _x , and _δt2 is the variance of x;

The operation process of the background difference: first, a background image frame B of the container above the conveyor belt is established by the method of mathematical modeling, and the current image frame is denoted as fn, and the gray values of the pixels corresponding to the background frame and the current frame are denoted as B(x , y) and f _n (x, y), according to the formula

Dn(x,y)=| _fn (x,y)-B(x,y)|

Subtract the gray values of the corresponding pixels of the two frames of images, and take their absolute values to obtain the difference image D _n :

Set the threshold T according to the formula

Binarize the pixels one by one to obtain a binarized image R _n ′. Among them, the point with a gray value of 255 is the foreground (moving target garbage) point, and the point with a gray value of 0 is the background point; the connectivity analysis of the image R _n ' is carried out, and finally the image containing the complete target can be obtained. garbage image R _n ;

This method can not only accurately locate the target to be recognized in the recognition process, but also can automatically intercept sample pictures from a large amount of picture information in a short time, which simplifies the process of deep learning to obtain samples. Caffe Convolutional Neural Network Framework

Caffe has a Python-related interface, which provides a deep learning architecture for the classification and image segmentation of the target garbage image in the present invention.

Lightweight Neural Network Model MobileNet

MobileNet can maintain a high accuracy in the task of garbage image recognition and classification.

The MobileNet model optimizes the traditional full convolution method, and decomposes the full convolution operation into Depthwise convolution and depthwise separable convolution, reducing the amount of parameters that need to be learned. On this basis, the MobileNet model sets two hyperparameters, the width factor and the resolution factor, to control the size of the model and the resolution of the input image, making the model more controllable in scale and speed of target garbage. When the width factor is constant, the resolution factor decreases, or the resolution factor is constant and the width factor decreases, the network parameters and multipliers will be relatively reduced; the introduction of the width factor α can obtain a smaller model with less computational loss . The function of α is to control the number of channels of garbage image input and output, the input channel is changed from M to αM, the number of output channels is changed from N to αN, and α is taken as 0.5. The resolution factor ρ can reduce the amount of calculation and parameters by ² times, allowing users to easily adjust the model. In the example, the value of ρ is 0.25, and the input resolution is set to 224, 192, 160 and 128.

The traditional convolution method undergoes 3×3 convolution, and then goes through the BN layer and the Relu activation function; while the depthwise separable convolution method used in the MobileNet algorithm: the 3×3 traditional convolution method is replaced by the Depthwise volume Product and 1×1 Pointwise convolution, and then go through BN and ReLU activation functions respectively like traditional convolution.

The MobileNet model consists of the following layers:

(1) Convolutional layer: The convolutional layer consists of multiple local filters, which are mainly used to extract different local features from the input target garbage feature map.

(2) Batch normalization layer: The batch normalization layer makes the whole model more stable and speeds up the training and convergence of deep convolutional networks.

(3) Scaling layer: The scaling layer scales and displaces the normalized neurons N. Since the invention uses the deep learning framework Caffe to train the MobileNet model and obtain parameters, the actual batch normalization is carried out in this framework. The unified calculation is divided into formula

Among them, mean, variance, scalefactor and ε are all learned parameters, mean is the mean vector of the same dimension as the feature map, variance is the variance vector of the same dimension as the garbage feature map, scalefactor is the dimension scaling factor, and ε is a very small The constant of , usually 0.0001;

(4) Nonlinear activation function layer: In order to make the lightweight neural network model MobileNet have nonlinear learning and expression capabilities, a nonlinear activation function layer is added to it. In the algorithm based on MobileNet, the nonlinear rectification function (ReLU) is used as the nonlinear activation function, and its calculation formula is as follows:

Specific implementation plan

After installing the Caffe deep learning framework and the MobileNet model, the sample images of the target garbage are trained to train the corresponding neural network. Taking a hazardous waste lighter as an example, changing the direction, position, and shape of the lighter to collect a data set multiple times, and upload the sample data set to the MySQL database.

The sample datasets for banana peels, soda cans, and batteries are obtained in the same way that a lighter is obtained by changing its shape. The sample dataset sizes used for training are shown in the table below.

Table 1: Sample Dataset Sizes

name Number of pictures/sheet lighter 460 banana peel 378 cans 355 Battery 407

The results of training and testing using the MobileNet model are shown in Table 2.

Table 2 Training and testing results

name test set/sheet Accuracy Fl-score lighter 150 97.52% 0.98 banana peel 150 96.23% 0.96 cans 150 98.19% 0.98 Battery 150 97.31% 0.97

Through testing, it is found that ordinary smart trash cans on the market cannot guarantee the recognition accuracy and recognition rate, and are often affected by the light environment. Due to the low accuracy rate and the low recognition rate, it leads to problems such as misclassification of garbage and low efficiency of garbage classification. This system uses the MobileNet model to solve the above problems very well. After consulting the literature, the data comparison between our invention and ordinary trash cans is shown in Table 3-5.

Table 3 Comparison data of ordinary smart trash can and the identification and classification rate invented by this team

name Ordinary trash can identification and classification rate Intelligent trash can identification and classification rate lighter 20.3s 5.1s banana peel 18.1s 4.5s cans 16.1s 5.4s Battery 16.3s 4.3s

Table 4 Comparison data of the recognition accuracy of ordinary smart trash cans and the invention of our team

name Ordinary trash can recognition accuracy Smart Trash Recognition Accuracy lighter 74.33% 97.52% banana peel 87.69% 96.23% cans 89.97% 98.19% Battery 80.31% 97.31%

Table 5 Comparative data of ordinary smart trash can and the invention of this team affected by light

light and darkness Ordinary trash can recognition accuracy Smart Trash Recognition Accuracy bright 83.71% 98.65% darker 72.94% 97.14% dark 54.36% 84.77%

It can be seen that the classification using machine vision and the MoblieNet model is significantly higher than the ordinary intelligent classification trash cans on the market in terms of recognition speed, recognition accuracy, and the ability to adapt to the environment. The experimental test shows that the system can be competent for the classification of 4 types of items including lighters, banana peels, cans, and batteries. The recognition accuracy is greater than 95%, and it has high practical application value in garbage classification.

Function realization of garbage conveying device

In order to free the hands of users, meet the needs of users' life, and create a fully automatic delivery mode, this product has developed an automatic garbage delivery device. The conveyor structure is shown in Figures 5, 7 and 8.

When the garbage identification and classification program finishes running, the corresponding value is returned to the main control system, and then the Raspberry Pi is used to communicate with the STM32 serial port, and the generated value is passed to the STM32. STM32 sends a command to the motor, and the motor drives the container containing the target garbage to move. After the container reaches the fixed position, STM32 continues to send commands to the steering gear to open the baffle at the bottom of the container and put the garbage into the corresponding type of trash can. The modules used in the specific transmission device are shown in Table 6.

Table 6 Conveyor Module Table

Other functions

In addition to identifying, classifying and automatically throwing target garbage, the present invention is also equipped with satellite positioning modules such as GPS to provide user location information for generating visual reports, odor deodorization, overflow detection, voice broadcast and other functions.

Table 7 Other function implementation schemes

Cloud Server

(1) Baidu cloud server

Operating system: centsos7.4

Hardware platform: single-core 2GHz CPU, 2GB memory

Support environment and version: python2.7, MySQL5.7

(2) Service background

The service background of the system uses python and sql programming, and uses the socket communication method to establish a long connection with the hardware to realize data transmission.

database

The database used in the delivery system is developed based on MySQL. For this system, the information that needs to be stored in the database of the server includes: the login account and password of the administrator; user information; the type and quantity of garbage in the trash can. So creating a database with three data tables can meet the needs.

The WeChat applet adopts the cloud development model and has its own database. For the client, the information in the database includes: account passwords of administrators and ordinary users; links to online news information; types of garbage; classification names; create four data tables, namely Can.

The applet calls the cloud function to realize the communication connection between the client and the MySQL database.

WeChat applet

For the convenience of users, this product uses WeChat applet to build the client, which can be easily acquired and disseminated in WeChat without installation, and has an excellent user experience.

The following functions can be realized through the mobile APP or WeChat applet:

(1) Triple real-time query method

Users can query the types of garbage through text search, image recognition, and dialect-equipped voice recognition, which greatly meets the needs of life.

(2) Intelligent recommendation function of classified information

The mini program will intelligently push various types of garbage classification news based on the user’s daily inquiries about the type and quantity of garbage, social trends and trends, and improve residents’ willingness and awareness of garbage classification.

(3) Unrecognized result upload function

Users can query the garbage that is not identified in the trash can on the applet, and upload the garbage pictures to the database if they are not found, and train the model independently through the system to achieve the purpose of correct identification next time.

(4) Intelligent robot online chat function

In order to achieve intimate companionship and entertainment purposes, the mini program client is also equipped with an intelligent chat interface, which can accurately judge the scene according to the user's words and send a response.

Intelligent garbage management function based on image recognition

(1) Garbage sorting automatic delivery function

This product can intelligently and automatically identify and classify the garbage thrown by the user. At present, it can be put into the corresponding trash can according to the four standards of dry garbage, wet garbage, recyclables, and hazardous garbage, without the need for users to manually classify, and meet the needs immediately. Solve classification troubles.

(2) GPS and other positioning functions

Determine the user's location and provide data for web crawlers to obtain home address information. According to the information provided by GPS, the output of different types of garbage in various areas of the city is analyzed, and information is collected for the visualization of management data.

(3) Odor deodorization function

Affected by the environment, the garbage may ferment in the barrel and produce peculiar smell. When the sensor in the intelligent sorting trash can detects the odor and reaches the threshold set by the odor sensor, it will send a signal to the main controller, and the main controller will supply power to turn on the ozone generation module to eliminate the odor. When the odor level returns to below the threshold, the power will be turned off. Ozone generator.

(4) Overfill reminder function

When the garbage reaches the specified height, it is considered to be overflowing. At this time, the main controller of the intelligent sorting trash can sends an overflow prompt signal to the applet to remind the user.

(5) Voice broadcast and key detection function

When the intelligent sorting trash can successfully identify the garbage, it will broadcast the type of garbage through the sound card, which can also improve the citizens' awareness of garbage sorting in the case of saving manpower. If the identification is unsuccessful, a voice broadcast will be used to remind the user that the garbage cannot be identified. At this time, the user can determine the type of garbage through the multiple inquiries of the WeChat applet. Small program photo upload.

Web version background management function

Visualization of garbage classification data: The present invention collects and analyzes the garbage classification and delivery data in the user's daily life, establishes a user garbage classification database, counts the type and quantity of garbage thrown by the user, and displays it in the form of visual icons, improving the convenience of management.

(1) Garbage truck travel time and route planning

The management terminal adopts big data, web crawler technology and GPS positioning function, uses user name, product number, etc. as keywords, crawls user garbage classification data on the Internet, and classifies the collected garbage classification data. Combined with cloud computing, the garbage overflow level of the community is estimated, and garbage trucks are notified to recycle garbage according to time. GPS satellite positioning technology is used to analyze the optimal route of garbage trucks, so as to timely recycle garbage in the community, avoid garbage accumulation, and facilitate the life of users.

(2) Planning of the waste incineration cycle of the landfill.

Claims (6)

1. an intelligent garbage classification system based on machine vision, is characterized in that, Including a garbage bin, which is provided with an image recognition garbage classification module, a temporary storage box and multiple sorting boxes; The image recognition garbage classification module includes MCU, trigger, camera and sub-throwing device; the trigger, camera and sub-throwing device are all connected with the MCU; The trigger is a pyroelectric infrared sensor, which is used to sense the human body approaching the trash can; The camera is used to obtain the image of the garbage in the temporary storage box; The MCU is used to perform image processing on the image and identify the type of garbage; The sorting device is used to send the garbage from the temporary storage box to the corresponding sorting box according to the sorting instruction issued by the MCU. 2. The intelligent garbage classification system based on machine vision according to claim 1, characterized in that, the sub-casting device comprises a conveyor belt (2), a pulley (3) and a conveyor belt drive mechanism for driving the conveyor belt to move; There are 2 pulleys, the conveyor belt is set on the 2 pulleys, the temporary storage box is fixed on the conveyor belt, the bottom of the temporary storage box is provided with a movable plate and a movable plate driving mechanism for controlling the opening and closing of the movable plate; the sorting box is located below the conveyor belt ; The sorting device also includes a position detection mechanism for detecting the position of each sorting box; Cooperate with the position detection mechanism, when the movable plate is opened, the garbage in the temporary storage box falls into a certain sorting box. 3 . The machine vision-based intelligent waste sorting system according to claim 1 , wherein the temporary storage box and the sorting box are provided with an odor sensor and an odor deodorization device. 4 . 4 . The intelligent garbage sorting system based on machine vision according to claim 1 , further comprising an overflow reminder module; the overflow reminder module comprises an ultrasonic sensor arranged in the sorting box for detecting the height of garbage. 5 . 5. The machine vision-based intelligent garbage classification system according to claim 1, further comprising a server; the MCU in the garbage bin is connected to the server through a communication module, and the operator can access the server through a PC or a mobile phone APP Get spam. 6. An intelligent garbage classification method based on machine vision, characterized in that, adopting the intelligent garbage classification system according to any one of claims 1-5; comprising the following steps: Step 1: Based on the trigger, detect the garbage thrower approaching the trash can; Step 2: Start image capture; After the garbage is put into the temporary storage box, start the image shooting to obtain the garbage image; Step 3: Garbage identification based on garbage images; Identify garbage based on the Caffe deep learning framework and the neural network MobileNet model; Step 4: classify the garbage based on the garbage identification result; Based on the garbage identification results, the identified garbage is transferred from the temporary storage box to the corresponding sorting box.

Images