- https://doi.org/10.1109/aisp68263.2025.11396162
An EfficientNet2-Based Automated Waste Classification System for Sustainable Resource Recovery
- Nov 22, 2025
- Shreyas Rajendra Hole +5 more
The ability to classify waste is fundamental to effective circular waste management; however, conventional visual sorting is time-consuming, expensive, and prone to errors. The proposed work addresses these issues by using automated convolutional neural networks (CNN) based systems with EfficientNetV2-B2 built on 15,000 annotated images over 30 classes and consolidated into 6 viable macro classes. The model generates 82.9% fine-grained accuracy, 93.4% macro accuracy, and 97.5% top 5 retrieval accuracy. The model, which is deployed with TensorFlow Lite, supports edge processing and can thus facilitate automated smart bins, automated recovery centres, and automated IoT-enabled recycling systems. It reduces human errors during waste sorting and is on track to provide energy-efficient and sustainable waste sorting systems to realise the vision of the circular economy.
