A deep learning-based method for efficient floating garbage debris recognition on high-performance edge computing platform

Our research introduces a novel method to identifying floating garbage debris using deep learning and High Performance Edge Computing (HPEC). We utilize a convolutional neural network (CNN) to classify debris from images captured by an RGB camera on a vessel, aiming for high accuracy and efficiency on low-power devices. We conducted a comparative analysis of various models, implementing optimization techniques like transfer learning and pruning. Each model was evaluated for accuracy, inference time, and energy consumption, leading us to develop a multi-objective function to determine the best approach. Our findings show that the proposed method effectively detects and classifies floating debris on desktop GPUs and low-power devices such as the Jetson Nano. Furthermore, it maintains accuracy by optimizing memory and computational power requirements while adapting to different energy needs. These advancements can enhance the capacity to combat marine plastic pollution and promote intelligent systems integration within environmental initiatives. They facilitate precise, real-time detection of floating debris on energy-constrained edge platforms, thereby effectively addressing deployment challenges in marine environments.