Garbage In, Insights Out: Deep Transfer Learning for Automatic Solid Waste Classification

Solid waste segregation is a significant global issue. Unlike advanced countries, where the solid waste segregation process is quite streamlined, this is not the case for developing countries, especially in those with large populations. India falls into the latter category and generates about 62 million tons of municipal solid waste annually, of which only about 20% is treated. This paper provides a comparative study of four competitive convolutional neural network (CNN) architectures (viz. ResNet50, EfficientNet, InceptionNet, and VGG16) to derive insights from waste-related data and classify them accurately into appropriate categories. These four CNN architectures were trained using a custom-made dataset consisting of 5392 images from 24 classes, covering the most commonly occurring categories of solid waste. The experimental results suggest that the proposed system can automatically segregate waste into appropriate categories, which can help develop a proper waste management system.