MTUSI developed a neural network for recognizing and sorting household waste
M.R. Menibaev, master’s student of MTUSI analyzed the problems of increasing and sorting household waste and developed a neural network configured to solve these problems. It is based on ResNet34 architecture and contains 34 disturbance layers. To train the neural network, a collective dataset was used, which is based on data located in the public domain and own images collected by the master’s student.
The size of the dataset is 2527 images of the main categories of household waste: glass, metal, plastic, cardboard, paper and several types of unsorted waste (mainly food). The classification accuracy of neural network objects is now 92.12%, says MTUSI.
Vyacheslav Voronov
Associate Professor of the Department “Intelligent Systems in Management and Automation” of MTSU, candidate of technical sciences, head of the Robotics Center of MTSU
“The good classification performance can be explained by the good structure of the neural network and its prior training, but the performance can always be improved by increasing the quality and number of images in the training set. Machine learning really makes it possible to qualitatively change the process of sorting garbage, which today is mainly implemented manually.”
The use of AI for waste sorting will increase the efficiency of waste processing, as well as have a positive impact on the environment, project participants are confident.
