Consumer-Grade Aerial-Aquatic Speedy Scanner (AASS) for Efficient Underwater Monitoring with Deep Learning

Abstract

Traditional marine monitoring techniques, such as line intercept transects, are often inefficient. While contemporary studies utilize autonomous underwater vehicles (AUVs) and the novel Portable Speedy Sea Scanner (P-SSS) to enhance marine monitoring, their widespread adoption is hindered by high costs and labor-intensive operations, with the data quality often failing to justify the investment. This study introduces a novel self-designed amphibious unmanned aerial vehicle (AASS), integrated with deep learning technology, which combines aerial and aquatic imaging capabilities. It surpasses traditional towed underwater survey tools and AUVs in ecological monitoring effectiveness. The research presents the AASS, enhanced with Super-Resolution Reconstruction (SRR) and an improved YOLOv8 detection network. The AASS system significantly improves data acquisition efficiency, capturing high-resolution images that precisely identify and categorize underwater targets. The SRR technique addresses issues like motion blur and low resolution, enhancing the YOLOv8 model's detection accuracy. Notably, the RCAN model achieved the highest mean average precision (mAP) of 78.6% for detecting reconstructed underwater images, with a x4 magnification in image quality showing superior mAP compared to the Bicubic method. These results affirm the efficacy of the proposed method in enhancing underwater ecological monitoring.