State-of-the-Art Review of Deep Learning-based Computer Vision Waste Detection: R-CNN, YOLO, Transformers, and Hybrid Models

Effective waste detection remains a pressing challenge due to the variability, clutter, and inconsistent visual characteristics of waste environments. This state-of-the-art review rigorously examines recent advancements in object detection models for waste detection and classification, covering the evolution from two-stage architectures (e.g., Faster R-CNN, Mask R-CNN) to single-shot detectors (YOLO family), recent Transformer-based models (e.g., ViT-WM, AL-DETR) and recent hybrid approaches. A structured literature review was conducted across major databases (2019-2025), guided by clearly defined inclusion and exclusion criteria. The review synthesizes reported results across studies to identify emerging trends, model capabilities, and deployment-readiness, while acknowledging the variations in datasets and evaluation conditions. Instead of ranking models, this review focuses on observed performance patterns and trade-offs. It also assesses the maturity of current techniques, analyzes limitations that hinder large-scale adoption, and highlights recent hybrid approaches and deployment requirements. Finally, it outlines future research directions toward building scalable, energy-efficient, and interpretable waste detection systems integrated with edge computing and smart infrastructure.