EcoSortAI: Deep Learning–Based Identification of Recyclable Materials and Contamination in Waste Streams

Recycling contamination remains one of the most significant barriers to efficient material recovery in modern waste-management systems. Improperly sorted waste introduces non-recyclable materials such as food residue, plastic films, and mixed waste into recycling streams, degrading material quality, increasing processing costs, and often causing entire loads to be diverted to landfills. Despite technological advancements, many recycling facilities still rely on manual sorting processes that are labor-intensive, inconsistent, and susceptible to human error. This study presents EcoSortAI, a deep-learning–based image classification system designed to automatically identify recyclable materials and detect contamination in waste images. A labeled dataset comprising 27 categories of recyclable and contaminant waste items was assembled and preprocessed using resizing, normalization, and augmentation techniques to simulate real-world variability. A MobileNetV2 convolutional neural network was trained using transfer learning to leverage pre-trained visual features while adapting to waste-specific classes. Model performance was evaluated using overall accuracy, per-class precision, recall, F1-score, confusion matrices, and multi-class ROC curve analysis. To assess practical feasibility, the trained model was converted to TensorFlow Lite format and deployed within a web-based application capable of real-time inference. The model achieved high classification accuracy and demonstrated strong generalization across diverse waste categories, including contamination classes. These findings indicate that lightweight deep-learning image classifiers can support automated waste sorting, reduce contamination rates, and improve recycling efficiency. EcoSortAI demonstrates the potential of scalable AI-driven environmental solutions that bridge the gap between academic research and real-world implementation.