Novosibirsk State Technical University (NETI) is working with a partner on an application that facilitates automatic garbage collection. A team of university programmers is developing a neural network for the application that can recognize the type and volume of waste, as well as an automated system for selecting a garbage carrier and a garbage disposal.
"NSTU-NETI's participation in the TRASH FOR CASH project began with a consultation on the creation of a platform for an online waste recycling service. We have joined the development of a single service that combines functionality for both the "producer" of garbage, whether it is an ordinary citizen or a legal entity wishing to get rid of waste and make money on it, and for the "consumers" of garbage in the face of processing enterprises in need of raw materials. There is already a portal where you can manually select all the necessary parameters and submit a request: the volume and type of garbage, the method and time of delivery, as well as the point of reception and disposal. Now students and I are actively working on creating a special module that will be able to automatically recognize the type of waste and estimate its approximate volume, which will facilitate the process for the user," said Alexander Yakimenko, Candidate of Technical Sciences, Head of the Department of Computer Engineering at NSTU-NETi.
To create a module with automatic garbage recognition, the development team decided to use a neural network. According to Alexander, the most difficult stage in the work is the training of a neural network based on the provided examples, in this case, photographs of waste of various classes. The "learning" process involves searching and uploading images of various wastes to a database, which are then analyzed and stored by a neural network, this action continues until the required accuracy is achieved in determining the content in the photo during subsequent upload. The system is also complicated by the variety of shapes and materials, for example, plastic and glass bottles are sometimes difficult to distinguish.
After a long process of preparation and selection of the necessary images, the first viable product was created. Next, there were several steps to configure and refine the recognition process, which led to the current workable version. At the moment, the neural network can determine with 98% accuracy the eight most common classes of garbage, including paper, plastic, glass, household waste and so on. In the future, the developers plan to expand the range of the neural network to at least 30 types of garbage.
For the user, the algorithm will be as follows: in the application, it is necessary to photograph waste, the neural network will process each photo, determine the type and volume and automatically enter all data into the application. The introduction of the recognition function into the current platform is planned in May.
The next stage of work on the project is a recommendation system that will help the user to find the waste receiving company in a couple of minutes and a car with a courier who will deliver the cargo to the final destination. This system is similar to the principle of operation of taxi applications: it is enough to take a photo of the waste, check the correctness of the data entered by the neural network, and click OK. At the specified time, the waste will be delivered to the enterprise for processing.
"Manufacturing enterprises, factories and shops dispose of waste every day. We are considering the possibility of creating an individual profile for regular customers, where statistics will be accumulated. This way we can further speed up and simplify the process of photographing and forming an application, since the user profile will already be linked to data on the types of garbage most often disposed of by the company, which will reduce the likelihood of recognition errors," added Alexander Yakimenko.