Ship Detection under the Influence of Raindrops in Rainy Environments Based on Integrated Generative Inverse Network and Improved YOLO Model

Ship detection plays a crucial role in intelligent ships. However, hip inspection on rainy days is a severe challenge. Raindrops can affect the visibility and stability of ship detection systems. To address these problems, we implement an integrated framework for accurate detection of target ships by transforming degraded images caused by rainy weather into clean images through adversarial neural networks. Firstly, we utilize Generative Adversarial Networks (GAN) to generate self-attention maps for focusing on the distorted areas in the degraded images. Secondly, a context compiler is employed to infer contextual information in the distorted areas, achieving high-quality restoration of these areas. Lastly, we introduce a Weighted Bidirectional Feature Pyramid Network (BiFPN) into the Yolov5 detection model to improve the detection and recognition accuracy of ships in distorted areas. The experimental results show that the framework can achieve excellent performance with average accuracy and average recall rates greater than 94%.