CN112722612A

CN112722612A - Garbage detection method and system based on YOLO network

Info

Publication number: CN112722612A
Application number: CN202011562683.1A
Authority: CN
Inventors: 何玄; 李清宏; 张大力; 拉凯什
Original assignee: Shanghai Jiaotong University; Xian Jiaotong University
Current assignee: Shanghai Jiaotong University; Xian Jiaotong University
Priority date: 2020-12-25
Filing date: 2020-12-25
Publication date: 2021-04-30

Abstract

The invention provides a garbage detection method and a garbage detection system based on a YOLO network, wherein the method comprises the following steps: acquiring garbage classification information of an area to be detected; the garbage classification information comprises a plurality of garbage types and wastes included in each garbage type; building a YOLO detection model according to the garbage classification information by using a YOLO network so as to build a classification vocabulary list for images corresponding to various wastes; and classifying the waste in real time according to the classification vocabulary table by using the YOLO detection model. By the aid of the method and the system, the garbage in the area to be detected can be automatically detected according to the garbage classification information of the area to be detected, so that real-time classification of the garbage is realized.

Description

Garbage detection method and system based on YOLO network

Technical Field

The invention relates to the technical field of image processing and garbage detection, in particular to a garbage detection method and system based on a YOLO network.

Background

With the development of urbanization, the difficulty of waste management is increased, and waste treatment methods are improved all over the world to improve treatment efficiency and enhance environmental protection. Such as waste collection (sensors that trigger alarms), intelligent logistics management software, autonomous cars, modern landfills, etc. Waste sorting is the first step in waste management where it is important to properly sort the waste on-site.

For example, the city municipal refuse classification program (TSP) policy entitled "Shanghai municipal domestic refuse management regulation" was officially enforced in Shanghai 1/7/2019. The Shanghai is the first city in China to implement the TSP policy to control the rapid growth of garbage. TSP policies involve three types of stakeholders: governments, enterprises and citizens. The relations and responsibilities of governments, enterprises and citizens should be clear, so that the classification of waste is gradually changed from mandatory to spontaneous to realize the mission of the normalization of the circular economy. In order to improve waste recycling and reduce the amount of waste, TSP policies have also been adopted in several growing big cities, such as beijing, chongqing, tianjin, hangzhou, guangzhou, chengdu, and sian. This policy is closely related to the daily lives of citizens and aims to change the behavior of citizens in disposing of waste.

However, each city has its own regulations on garbage disposal, transport and fines. In Shanghai, four categories can be classified: recyclables, hazardous waste, dry waste and wet waste; in Beijing, the garbage can be classified into four types of recyclable materials, dangerous garbage, kitchen garbage and other garbage; chongqing is classified as recyclables, hazardous waste, perishable waste and other waste. The other 43 cities mostly adopted the Beijing and Chongqing categories. In fact, although the categories are distinguished, the meanings of the main categories are similar. In Shanghai, the leading sheep and the demonstration city of the Chinese TSP policy, large amounts of waste, decorative waste and electronic waste are further categories of business topics. Furthermore, there are 104 spam items as a guide and example for the four spam categories. As a mandatory policy for garbage classification, individuals who do not want to be classified correctly after passing persuasion are penalized with high fines. In Shanghai, a fine of 200 Yuan or more will be made; in Beijing, individuals are charged a fine of 50 to 200 dollars.

Despite the punishment, the accuracy of the food waste is only over 20% as exemplified by the statistics of Beijing. In addition, the garbage classification confusion between cities is also very serious. According to the latest comments of public opinions of Chinese TSP published in 2019, 325,748 social public opinions from WeChat public accounts and Xinlang microblog are analyzed to obtain positive, moderate and negative opinions, wherein the positive opinions with the highest percentage indicate that the public supports the environmental protection policy. However, in view of the negative emotion of public opinion, countermeasures and suggestions have been determined from three aspects: 1) a support facility; 2) guiding the thought; 3) the implementation is ensured. In particular, negative sounds include: 1) the process of facilitating the sorting of waste is too hasty; 2) the forecast is not enough; 3) infrastructure is incomplete; and 4) the delivery process is complicated. The cumbersome and complex process of garbage classification has become a key problem for citizens to drive the need for innovative solutions. Therefore, there is a need for technical programs to assist municipalities in managing waste. Furthermore, it is estimated that by 2050, more than 50% of the global population will live in cities in asia-pacific regions, and therefore, related innovations in waste management have become an important topic of discussion.

Similarly, in india, the federation environment division recently announced a new 2016 "solid waste management act (SWM)". New regulations require source classification of waste in order to convert it into wealth through recycling, reuse and recycling. Waste producers must classify waste into three categories: biodegradable, recyclable (plastic, paper, metal, wood, etc.) and hazardous waste (diapers, napkins, mosquito repellants, cleaners, etc.) that is then handed over to waste collectors (a significant person in the indian waste management framework). 2016SWM is being customized in the main areas of Delri, Mony, etc. It can be seen that the indian waste management system also varies greatly in its manner of operation, taking about 4 to 5 years.

Waste classification is the process of classifying waste, i.e. refuse, into different elements. Sorting involves varying degrees of manual and automated work. In asia, the classification of waste in cities has traditionally relied on informal waste collectors (important participants) in the waste supply chain. However, with the increase in urban living costs, the decrease in prices of waste products, and the decrease in welfare policies, there is now a need for greater use of automation to maximize waste recovery and prevent landfill. Therefore, there is a need in the art for an automated waste detection system that can classify waste in real time, maximize waste recovery, and prevent landfill.

Disclosure of Invention

The invention aims to provide a garbage detection method and system based on a YOLO network, which can be used for automatically detecting garbage in a to-be-detected area according to garbage classification information of the to-be-detected area so as to realize real-time garbage classification.

In order to achieve the purpose, the invention provides the following scheme:

a garbage detection method based on a YOLO network comprises the following steps:

acquiring garbage classification information of an area to be detected; the garbage classification information comprises a plurality of garbage types and wastes included in each garbage type;

building a YOLO detection model from the garbage classification information using a YOLO network to build a classification vocabulary for images corresponding to various wastes,

and classifying the waste in real time according to the classification vocabulary table by using the YOLO detection model.

Optionally, the step of classifying the waste in real time according to the classification vocabulary by using the YOLO detection model includes:

acquiring an image to be detected;

and inputting the image to be detected into the YOLO detection model according to the classification vocabulary, and outputting a detection result.

Optionally, the inputting the image to be detected into the YOLO detection model according to the classification vocabulary, and outputting the detection result further includes:

and loading the detection result to a display device.

Optionally, after the loading the detection result to the display device, the method further includes:

rewarding the operator according to the classification condition of the operator on the waste; wherein the operator classifies the waste according to a display result on the display device.

Optionally, the method further includes: the operator pays the user who discarded the trash according to the discarded trash.

Optionally, the acquiring the garbage classification information of the area to be detected specifically includes:

acquiring the position of a user to be detected;

determining the current garbage management policy document of the area to be detected according to the position of the user to be detected; the area to be detected is the area of the position of the user to be detected;

and determining the garbage classification information of the area to be detected by adopting a probabilistic word clustering method according to the garbage management policy document.

Optionally, the detection result includes whether the image to be detected contains a garbage target, and a specific position and a category of the garbage target.

A garbage detection system based on a YOLO network comprises:

the garbage classification information acquisition module is used for acquiring garbage classification information of an area to be detected; the garbage classification information comprises a plurality of garbage types and wastes included in each garbage type;

a YOLO detection model building module for building a YOLO detection model from the trash classification information using a YOLO network to build a classification vocabulary for images corresponding to various trash,

and the real-time classification module is used for classifying the wastes in real time according to the classification vocabulary table by using the YOLO detection model.

Optionally, the real-time classification module includes:

the image acquisition unit to be detected is used for acquiring an image to be detected;

and the detection result output unit is used for inputting the image to be detected into the YOLO detection model according to the classification vocabulary and outputting a detection result.

Optionally, the garbage detection system based on the YOLO network further includes:

and the loading module is used for loading the detection result to the display equipment.

the rewarding module is used for rewarding the operator according to the classification condition of the operator on the waste; wherein the operator classifies the waste according to a display result on the display device.

and the fee payment module is used for paying the fee to the user who discards the garbage according to the discarded garbage by the operator.

Optionally, the garbage classification information obtaining module specifically includes:

the position acquisition unit is used for acquiring the position of a user to be detected;

a policy file determining unit, configured to determine a current garbage management policy document in the to-be-detected area according to the location of the to-be-detected user; the area to be detected is the area of the position of the user to be detected;

and the classified information determining unit is used for determining the garbage classified information of the area to be detected by adopting a probabilistic word clustering method according to the garbage management policy document.

According to the specific embodiment provided by the invention, the invention discloses the following technical effects: in the garbage detection method and system based on the YOLO network, the garbage classification information is extracted aiming at the to-be-detected area so that the classification result is more accurate due to the fact that the garbage classification information of different areas is different; the classification of garbage using the YOLO network model enables automated real-time classification results.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.

Fig. 1 is a flowchart of a garbage detection method based on a YOLO network according to embodiment 1 of the present invention;

fig. 2 is a schematic structural diagram of a garbage detection system based on the YOLO network according to embodiment 2 of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.

Fig. 1 is a method for detecting spam based on a YOLO network according to embodiment 1 of the present invention, as shown in fig. 1, the method includes the following steps:

step 101: acquiring garbage classification information of an area to be detected; the garbage classification information includes a plurality of garbage types and wastes included in each of the garbage types.

Step 102: building a YOLO detection model from the garbage classification information using a YOLO network to build a classification vocabulary for images corresponding to various wastes,

step 103: and classifying the waste in real time according to the classification vocabulary table by using the YOLO detection model.

In the embodiment of the present invention, step 101 specifically includes:

step 11: and acquiring the position of the user to be detected.

Step 12: determining the current garbage management policy document of the area to be detected according to the position of the user to be detected; and the area to be detected is the area of the position of the user to be detected.

Step 13: and determining the garbage classification information of the area to be detected by adopting a probabilistic word clustering method according to the garbage management policy document. The garbage classification information includes a plurality of garbage types and garbage included in each garbage type.

Since the garbage management policies of the regions are different, the garbage management policy of the user position can be determined by detecting the user position so as to realize the most accurate classification. In the embodiment of the present invention, the location of the mobile phone of the user (i.e., the location of the user) may be identified by using a GPS technology in the mobile phone, so as to determine the area to be detected, but is not limited thereto. Then, the corresponding garbage management policy document is determined according to the determined region to be detected, and the determination is completed manually or automatically. Wherein the automatic manner may feed spam management policy documents for systems within a range of locations set by a system administrator or crawled over a network. In addition, it is also possible to provide support in different languages for people from different countries/regions depending on where the user is located, or to display targeted information to visitors from different locations.

The method for clustering probability words in the embodiment of the present invention is also called "Excessive Topic Generation (ETG)", and its specific steps and formulas are disclosed, specifically, see the article "Intelligent collaborative patent mining using Excessive Topic Generation". The probabilistic word clustering method is a keyword extraction method based on a topic model. The method comprises the steps of extracting keywords from a garbage management policy document representing relevant information of waste classification by adopting a probabilistic word clustering method, specifically, clustering text corpora in the garbage management policy document according to the number of garbage types to obtain keywords (namely waste) included under each garbage type, wherein the clustering value is set as the number of the garbage types. In the embodiment of the invention, the number of the garbage types is assumed to be n, so that the clustering value K is set to be n in the K-means cluster, n groups of clustering results can be obtained by the method, each cluster represents one garbage type, and the keyword in each clustering result represents the waste contained in the garbage type. The garbage types of the area to be detected and the wastes included by each garbage type can be accurately obtained through the steps.

For example, in shanghai, china, the number of garbage types is 4, and 4 groups of clustering results, namely recyclables, dangerous garbage, dry garbage and wet garbage, can be obtained by the method; wherein, the recyclables can be represented by blue, the dangerous waste can be represented by red, the dry waste can be represented by blue and the brown, and the wet waste can be represented by black. The keywords in each cluster result represent the waste included in the garbage category, and the keywords of recyclable household garbage such as waste paper, waste plastics, waste metals, waste fabrics, and the like in the recyclables cluster result belong to the garbage in the recyclables. For another example, in india, if the number of garbage types is 3, 3 groups of clustering results can be obtained by this method, i.e., the garbage types can be classified into biodegradable garbage, non-biodegradable garbage and dangerous garbage, wherein the biodegradable garbage can be represented by green, the non-biodegradable garbage can be represented by blue, and the dangerous garbage can be represented by black. The keywords in each clustering result represent the wastes included in the garbage category, for example, the keywords in the dangerous garbage clustering result, such as diapers, napkins, mosquito repellents, detergents, etc., belong to the wastes in the dangerous garbage.

Step 102: a YOLO detection model is built from the garbage classification information using a YOLO network to build a classification vocabulary for images corresponding to various waste. In the embodiment of the present invention, the YOLO network may adopt a YOLO V3 framework.

The training data employed in the embodiments of the present invention was a summary from an open source image dataset, specifically a large dataset created by MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) that contained 187,240 images, 62,197 annotated images and 658,992 tagged objects. And establishing a YOLO detection model and a classification vocabulary according to the training data, a YOLO V3 framework and the garbage classification information of the to-be-detected region.

The basic principle of YOLO is: firstly, dividing an input image into grids of S multiplied by S, predicting 5 bounding boxes for each grid, then removing a target window with low possibility according to a threshold value, and finally removing a redundant window by using a bounding box combination mode to obtain a detection result. The building process of the YOLO detection model specifically comprises the following steps: performing frame marking on target garbage on the image by using a marking tool to classify the garbage as training data, wherein the color of a boundary frame represents the category of the target garbage, and the training data comprises a training set, a test set and a verification set; and calling training data to train the source code of the YOLO V3 framework, and obtaining a YOLO detection model after the training is finished. In the process of labeling the images, the image content in the boundary box and the color of the boundary box are called classified vocabularies, and the labeled images are collected to obtain a classified vocabulary. And later, when garbage detection is carried out on the region, the garbage detection can be finished according to the classification vocabulary. Here, the classification vocabulary includes images corresponding to various wastes, classifications to which the wastes belong, and colors of boxes corresponding to the classifications.

It should be noted that, due to different garbage management policies of the respective regions, the classification vocabularies of the different regions are also different, and therefore, when detecting garbage in the different regions, the classification vocabularies of the regions to be detected need to be used.

In the embodiment of the present invention, step 103 specifically includes:

step 31: acquiring an image to be detected;

step 32: and inputting the image to be detected into the YOLO detection model according to the classification vocabulary, and outputting a detection result.

And the detection result comprises whether the image to be detected contains the garbage target or not and the specific position and the category of the garbage target. Specifically, the image to be detected may include a plurality of spam targets, and the position and type of each spam target may also be different, so that all spam targets in the image and the positions and types of spam targets can be detected respectively according to needs, and a certain type of spam can be detected according to needs.

For example, in the embodiment of the present invention, a plurality of targets are selected according to the classification vocabulary, for example, when a certain type of garbage in the area is detected, all images of the type of garbage are selected from the classification vocabulary, a garbage target is detected from the image to be detected, a position of the garbage target is described by using a bounding box, and a type of the garbage is represented by using a color of the bounding box. The bounding box is a rectangular box and can be determined by the x-axis coordinate and the y-axis coordinate of the upper left corner of the rectangle and the x-axis coordinate and the y-axis coordinate of the lower right corner of the rectangle; the color of the bounding box is established based on the sorted vocabulary and thus matches the garbage management policy of the location of the user. For example, in the indian area, assuming that apples and plastic bottles are detected in the image, a green frame is formed around the apples because it belongs to the biodegradable category; on the other hand, the plastic bottle will have a red frame around it because it belongs to the category of non-biodegradable; finally, the garbage target will be marked as garbage category specific to the region (depending on the garbage management policy of the region to be detected).

After step 103, the method may further comprise: and loading the detection result to a display device. For example, the detection result may be loaded into an Augmented Reality (AR) framework (android/ios/unity game engine) according to the specification of the display device, and the real-time classification information may be displayed on the display device in the form of a bounding box or other relevant classification scheme.

The method may further comprise: rewarding the operator according to the classification condition of the operator on the waste; wherein, if the operator classifies the waste according to the display result on the display device, the operator can be given a certain reward, and the reward can comprise points, coupons and the like. The purpose of this step is to serve as a platform for deploying value-added services (the display result can be output as an Application Programming Interface (API) to build the value-added services on the basis of garbage classification). These value added services may include location based gaming, mobile advertising, mobile money, mobile commerce, mobile publications, and the like.

The method may further comprise paying a fee by the operator to the user who discarded the garbage based on the discarded garbage. Specifically, as an alternative embodiment of the value added service, a user who discards the garbage may be paid a fee according to the discarded garbage by an operator, such as a garbage collector (informal actor in the garbage management system). The garbage collector will then resell these items to a recycling center and profit from his investment. In this case, a value added service may be provided, which the value added service platform will perform visual classification during recycling to estimate the total amount to help the user estimate as early as possible the tentative fee that he will charge when he discards non-degradable recyclable items. However, this is not the only value added service, but there are many area customizations and value added possibilities of the proposed invention. In addition, in the chinese system, the exchange of coupons or the addition of value through games can be performed.

An embodiment 2 of the present invention provides a garbage detection system based on a YOLO network, and as shown in fig. 2, the system includes:

a garbage classification information obtaining module 201, configured to obtain garbage classification information of a to-be-detected area; the garbage classification information includes a plurality of garbage types and wastes included in each of the garbage types.

A YOLO detection model building module 202, configured to build a YOLO detection model according to the garbage classification information using a YOLO network, so as to build a classification vocabulary for images corresponding to various wastes.

A real-time classification module 203, configured to classify the waste in real time according to the classification vocabulary using the YOLO detection model.

As an optional implementation, the real-time classification module 203 includes:

and the image to be detected acquiring unit is used for acquiring an image to be detected.

As an optional implementation manner, the garbage detection system based on the YOLO network further includes:

As an optional implementation manner, the garbage classification information obtaining module specifically includes:

and the position acquisition unit is used for acquiring the position of the user to be detected.

A policy file determining unit, configured to determine a current garbage management policy document in the to-be-detected area according to the location of the to-be-detected user; and the area to be detected is the area of the position of the user to be detected.

As an optional implementation manner, the detection result includes whether the image to be detected includes a garbage target, and a specific position and a category of the garbage target.

According to the garbage detection method and system based on the YOLO network, the wastes in the area to be detected are classified in real time, the classification information is presented as augmented reality, and then the augmented reality conversion is also presented as a platform for deploying value-added services. Compared with the prior art, the method has the advantages that key information can be extracted from the garbage management policy document, an internal classification group, namely a classification vocabulary, is created, and the classification result can be presented in augmented reality; the deployment of the value added service can enable a plurality of participants in the forms of enterprises, industries, communities, families, individuals and the like to better develop activities of garbage management and resource recovery, maximize garbage recovery and prevent garbage landfill.

The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The system disclosed in embodiment 2 corresponds to the method disclosed in embodiment 1, so the description is simple, and the relevant points can be referred to the description of the method.

The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.

Claims

1. A garbage detection method based on a YOLO network is characterized by comprising the following steps:

building a YOLO detection model according to the garbage classification information by using a YOLO network so as to build a classification vocabulary list for images corresponding to various wastes;

2. The YOLO network-based spam detection method of claim 1, said step of real-time classifying trash according to the classification vocabulary using the YOLO detection model comprising:

acquiring an image to be detected;

3. The method as claimed in claim 2, wherein the inputting the image to be detected into the YOLO detection model according to the classification vocabulary, and outputting the detection result further comprises:

and loading the detection result to a display device.

4. The method of claim 3, wherein the loading the detection result onto a display device further comprises:

5. The YOLO network-based spam detection method of claim 1, wherein the method further comprises:

the operator pays the user who discarded the trash according to the discarded trash.

6. The method for detecting spam based on the YOLO network of claim 1, wherein the obtaining the spam classification information of the area to be detected specifically comprises:

acquiring the position of a user to be detected;

7. The method as claimed in claim 2, wherein the detection result includes whether the image to be detected contains a garbage target, and the specific location and category of the garbage target.

8. A garbage detection system based on a YOLO network is characterized by comprising:

9. The YOLO network-based spam detection system of claim 8, said real-time classification module comprising:

10. The YOLO network-based spam detection system of claim 9, further comprising:

11. The YOLO network-based spam detection system of claim 10, further comprising:

12. The YOLO network-based spam detection system of claim 8, further comprising:

13. The system of claim 8, wherein the garbage classification information obtaining module specifically comprises:

14. The method as claimed in claim 9, wherein the detection result includes whether the image to be detected contains a garbage target, and the specific location and category of the garbage target.