Artificial Intelligence Applied to waste classification

Currently, waste management and classification are one of the biggest problems our society is facing. The waste generated in the population centres is bigger and bigger, which makes the problem even worse. Recycling and reuse are the aims of the solution to it. However, when recycling diverse waste it is necessary to keep a correct separation of the different types existent so they can be treated correctly. Owing to the complexity of this problem, the use of artificial intelligence to waste classification together with Industry 4.0 technologies allows for improving the results and the efficiency of these processes.

Waste classification with artificial intelligence

The growing need to reduce waste generated makes it imperative to recycle all possible items, thus it is necessary to separate de different kinds of waste, which allows carrying out the appropriate treatment with each one. Despite having specific bins for recycling, in those where waste is separated before reaching the plant, mistakes are inevitably made. Also, there is a high number of people that do not recycle yet.

Traditionally, in recycling plants, the work of separation has been carried out manually by people who, apart from separating those wastes, were in charge of classifying the different types in bins and smart containers that, through artificial intelligence, classify wastes according to how they are discarded. Although for now, this kind of classification does not give as good results as the applicated in the recycling plants themselves.

The classification process with artificial intelligence

To be able to apply artificial intelligence to waste classification it is necessary to do several previous steps until achieving a fully functional station.

For start, it is recommended that the dates capture be done with the station which is intended to remain installed finally. Thus, there will be few differences between the testing process and the real one, saving a lot of time. In this station will be necessary to adjust the camera´s capture parameters and determine the lighting to use, as well as the dates to obtain.

Also, it is necessary a large number of images (thousands of them) with which the artificial vision model can be trained. When a model is trained, what we do is show them the objects that we want that detect in the image (thus obtaining his position) and how we want them classified. For this reason, thousands of images are needed, since the model must learn how to recognize all possible objects.

Technology used for waste classification with artificial intelligence

When artificial intelligence is used for waste classification diverse technologies are used, both for the process of the classification itself and for what is desired to do subsequently with the data obtained.

Robotic arm

If a completely automatized station is desired it will be necessary to install a robotic arm. Its mission will consist of receiving the obtained information in the classification process by the AI and using these data for separating the diverse types of waste detected.

Cameras

The artificial vision cameras are needed to obtain the data. The usual thing is that they are placed on the conveyor belt on which the waste passes to capture the images that subsequently will be used to train the model and detect waste. Normally, 2D-ready cameras are used for this kind of installation. However, depending on the type of data you want to analyse, it is possible to use other types of cameras more complex, such as 3D or hyperspectral.

Deep Learning

Talking about artificial intelligence, especially for a system as complex as the classification and detection of waste, habitually we refer to Deep Learning. These algorithms of artificial intelligence use neural network models, which are based on human brain behaviour and currently are the algorithms more used for the applications of artificial vision with IA due to their good performance and good results.

Lighting

When obtaining images, it is necessary that, to achieve the best possible results, these have a series of characteristics, like there is little variability between photos no matter the time of day, or that all the elements can be perfectly visible. For that will be necessary to choose correctly the kind of lighting to be used together with the cameras.

Benefits

When applying artificial vision for waste classification, more efficient treatment is achieved. Thus, it is possible to do the separation and classification of a bigger amount of waste in much less time than a normal person would need manually.

Additionally, the application of IA systems for the classification, although they need an initial investment, represents cost savings in the long term, contributing, also, to circular economy and environment protection.

Fulfilled Projects

In ATRIA artificial intelligence projects have been done to classify the waste, such as SIARA and PERTE.

SIARA

In this project, the waste classification was done through the methods previously described. Using a camera on the conveyor belt to take the pictures, various residues were located and classified in these images, particularly importance was given to bulky plastics that, habitually, give more problems and must be processed differently.

Concretely was essential to locate residues such as plastic films or nacelles used for fruit transport. However, although these were the most important residues, it was capable of localized cardboard waste and electronics.

SEPARA

Similar to SIARA, residues are located on a conveyor belt. But, in this case, other technologies and tools were used in addition to the vision camera.

To use artificial intelligence for residue classification, it is necessary a large quantity of data. For this reason, tools that could add new information have been used in this project, such as a linear camera, a 3D sensor and a hyperspectral vision camera.

The lineal camera is similar to a conventional vision camera, only that is capable of capturing bigger size pictures with higher quality. In this case, it was used to take a picture of the entire stretch of tape where there was debris, that is to say, you could see a whole stretch of residue without cuts, avoiding like this that information was between one image and the next one.

The 3D sensor allows taking data about the size and height of the residue. Thus, it is possible to know how voluminous the waste is and how to treat it.

A hyperspectral camera is a special sensor that allows knowing the composition of the image´s objects. In this way, the different types can be identified, such as plastics or cardboard, and so be able to treat them how it would be necessary.

Do you want to apply Artificial intelligence for the waste classify in any of your projects? Contact with us!