EcoBin: A Two-Stage Deep Convolutional Neural Network for Contamination-Aware Waste Classification
Researchers have developed EcoBin, a novel two-stage deep convolutional neural network designed to improve waste classification by accounting for contamination in recyclables. The first stage, built on an EfficientNetV2-S backbone, categorizes waste into disposal pathways, while the second stage specifically identifies and flags contaminated items destined for recycling. To address the lack of public data on contaminated recyclables, a synthetic dataset was created using U2-Net for segmentation and realistic contamination textures. The complete EcoBin pipeline demonstrated significant improvement, correctly routing 24 out of 25 contaminated items, a substantial increase from the base classifier's 1 out of 25. AI
IMPACT This research could lead to more effective automated waste sorting systems, reducing contamination and improving recycling efficiency.