In this work, a total solution is proposed for a typical underwater drone with high quality and low power consumption dedicated to inshore aquaculture. The proposed method introduced a high-perceptibility underwater image correction algorithm using a structural similarity evaluation to improve the video quality and object detection accuracy. The proposed algorithm considered real-time performance and low-power consumption by redesigning a compact model. It also employed automatic mixed precision (AMP) to effectively reduce the computational redundancy. An efficient implementation is also introduced using a low-cost GPU board, namely Jetson Xavier NX. The proposed implementation demonstrates improved object detection performance with a processing speed up to 60 fps.