A vision-based litter detection and classification using SSD MobileNetv2

Municipal solid litter generation has steadily increased significantly in recent years. To prevent waste from collecting in the environment, it is beneficial to clean up wastes and rubbish using autonomous cleaning equipment such as unmanned surface vehicles. Cleaning efficiency requires a high-accuracy and reliable object detection system. Many machine learning techniques are explored so far to produce litter recognition systems. There are few studies on the effectiveness of cuttingedge deep learning object detection algorithms in the domain of litter localization and detection. We propose a detection and classification module for five different waste objects in this research. The dataset is created by gathering photos from various sources and under various environmental conditions at our Institute. In order to enhance the quantity of data, data augmentation techniques are used. Image processing and feature extraction steps are applied to enhance the performance of the proposed model. The SSD MobileNetv2 model is employed for litter recognition which detects and identifies distinct contaminants and hazardous waste materials with a mean Average Precision (mAP) of 0.84. The proposed module aids in environmental cleanup by efficiently detecting the waste objects present in the surrounding environment.