Analysis of using Zero Shot Open Vocabulary Detection Methods for Plastic Waste Classification

Pollution from plastic garbage has grown to be a major environmental issue, requiring the creation of effective techniques for identifying and categorizing it. Supervised machine learning is a proven technique for automatic waste detection and classification but requires heavy training. Unsupervised learning techniques such as transfer learning and few-shot learning (FSL) have greatly reduced the training time and the need for huge datasets. Vision Language Models (VLMs) with open vocabulary and zero-shot detection techniques have reduced the shortcomings of traditional models. This paper analyzes the scope of using zero-shot combined with open-vocabulary detection techniques such as YOLO World, Grounding DINO, ViLD, and OWL-ViT for plastic waste detection. Zero-shot open vocabulary models eliminate the need for training custom data for long hours, enabling models to classify new classes with only their semantic encoders and contrastive pre-training.