Waste segregation at the source plays a pivotal role in effective recycling and sustainable waste management. However, manual sorting is labor-intensive, error-prone, and often inefficient, especially in urban environments with high waste volumes. This research proposes an AI-driven, automated waste classification system that accurately classifies waste into two primary categories: biodegradable and non- biodegradable, enabling better downstream processing such as composting, recycling, and landfill management. We explore and evaluate several state-of-the-art deep learning architectures, focusing on lightweightandoptimizedconvolutionalneuralnetworks (CNNs) suchasMobileNetV2, EfficientNet-B0,and ResNet-50, along with object detection models like YOLOv8 for real-time waste stream applications. Given the constraints of edge devices (e.g., smart bins) model optimization strategies including quantization, pruning, and andknowledgedistillationtoreducemodelsizeandinferencetimewithoutcompromisingaccuracy. These optimizations enable deployment on low-power hardware like Raspberry Pi, NVIDIA Jetson Nano, and other IoT-compatible devices Keywords: Automated Waste Classification, Convolutional Neural Networks, MobileNetV2, EfficientNet-B0, ResNet-50, YOLOv8, IoT Devices, Raspberry Pi, Jetson Nano, Real-Time Object Detection.
Selvamalar et al. (Tue,) studied this question.