Development of A Deep Learning Hybrid Model for Classification of Plastic Waste Objects in the Coastal Environment

Plastic, glass, metal, and wood waste has become a global scourge and nightmare to the marine environment and the development of the coastal areas. This paper aims at developing a hybrid model of deep learning technique using Convolutional Neural Networks (CNN) and Vision transformers (ViT) in the performance of detecting underwater waste materials. The model is designed with the use of transfer learning and high-resolution imaging sourced from underwater drones to help overcome the challenges of underwater work such as poor visibility and insufficient ambient light. The local features are extracted using CNNs and the global spatial relationships of the images are attended to by the ViTs. The results achieved a waste detection accuracy rate of about 90%, which is effective even in difficult environmental conditions. The concept is for a waste management system where drones or underwater robotics systems can assist automated waste management, waste treatment processes, and aid marine renovation activities.