Deep Learning-Driven Real-Time Visual Pollution Detection and Multi-Class Waste Classification in Urban and Textile Landscapes

Abstract

The expanding issue of visual pollution in urban and textile environments has a negative influence on public health, aesthetics, and overall quality of life. Unmanaged garbage, crowded regions, and poorly maintained structures are some prevalent factors that reduce the aesthetic appeal of cities and industrial areas. The growing problem of visual pollution in urban and textile contexts is the subject of this study. As urbanization and industrial activity rise, sources of visual pollution including uncontrolled garbage, urban congestion, and badly kept industrial zones become increasingly noticeable. Various rubbish categories are classified by the waste sorting system in a publicly accessible dataset. This paper suggests a deep learning-based system for monitoring and identifying visual pollution in real time utilizing cutting-edge deep learning techniques, using developments in deep learning and real-time image identification. The objective of this research is to develop a predictive model for the automatic classification of thirty distinct waste object kinds using SSD (Single Shot Multibox Detector) technology. There are five pre-trained networks as backbone networks for feature extraction: ResNet-50, ResNet-18, MobileNetV3, and EfficientNetV5.When compared to the current conventional approaches, the applied model performs better for classification(ResNet-18) in terms of accuracy 96%, precision 96.3%, recall 95.83%, and F1-score 96.13% and for detection(SSD MobileNet V2) in terms of precision 98.7%, recall 98.5%, and mAP50 98%. The findings imply that the system can successfully increase the efficacy and precision of garbage sorting procedures in practical situations.