Waste Segregator: An Optimized Neural Learning Approach Towards Real-Time Object Classification

Abstract

In the present era, natural components such as air, water and land tend to act as pollutant collectors. On the road to proper waste management, human beings have developed multiple methods, and most of these require the categorical separation of waste at the earliest possible stage. This paper on a smaller scale focuses on identifying and classifying the waste into two broad categories, recyclable and organic. We used a dataset that contains nearly 22,000 images including all the classes. The model used for the work was convolutional neural network (CNN) and can be optimized more by finding optimal hyperparameters. The CNN consists of two convolutional layers followed by max pooling, fully connected layer, and an efficient GPU to make the training faster. An accuracy of 92% and above was observed with the given dataset using CNN. The experiments conducted prove that CNN is more efficient than support vector machines (SVM); however, SVM tends to give better results on smaller datasets, depending on how the data is partitioned for training.