YOLOv5 Water Surface Garbage Detection Algorithm Based on Deep Separable Convolution

The challenges posed by multiple parameters, high computational complexity, and a single convolutional kernel in YOLOv5 lead to subpar accuracy and slow detection rates when applied to water surfaces. To address these issues, we propose a novel YOLOv5 water surface target detection algorithm based on deep separable convolution. This algorithm employs deep separable convolution to reconstruct the Darknet53 network, effectively reducing computational complexity. Additionally, we integrate the Inception network structure into the detection network, elevating both complexity and detection accuracy. Evaluation of the algorithm on the dataset reveals outstanding performance, boasting an impressive 95.3% accuracy and a rapid detection rate of 65.34f/s. These results significantly surpass the original YOLOv5 water surface garbage detection algorithm, demonstrating the efficacy of our proposed method.