A Lightweight Deep Neural Network-Based Optical Image Detection Method for Underwater Submerged Rocket Wrecks

Maritime recycling plays a significant role in reusing components from spacecraft. However, the current methods for detecting and recovering sunken spacecraft wrecks suffer from inaccurate detection and low efficiency. To improve the detection effectiveness of rocket wrecks, this paper proposes a method based on underwater optical images which leverage the high-resolution features of images. To address the issue of low accuracy and efficiency in detecting rocket wrecks, a lightweight deep neural network detection model is introduced to enhance accuracy while reducing the computational load and increasing efficiency. Finally, through quantitative and qualitative analyses, our proposed method achieves a detection accuracy of 96.0% with low computational power, which lays a solid foundation for spacecraft recycling.