Haze Removal And Ship Tracking In Hazy Weather Based On Deep Learning

When the tracking algorithm is applied to a hazy maritime image with some small ships, the algorithm performance is negatively affected. To address the above problems, a framework for ship tracking in hazy weather using deep learning methods is proposed in this study. First, a lightweight convolutional neural network (CNN) is used to remove haze from images. Second, the YOLOv5 algorithm with a small target detection layer added is used to detect ships in dehazed marine images, and the Deep SORT algorithm is used to track ships. Then, multi-target tracking evaluation metrics are used to evaluate the tracking effectiveness of ships in videos taken in hazy environments and ships in videos after removal of haze. The experiments show that the PSNR of the 0.96, indicating that the framework can effectively improve the quality of hazy maritime images. The evaluation metrics for multi-objective tracking also show that for hazy videos with small ships, the ship tracking algorithm is more robust when applied after haze removal.