A ship object detection algorithm based on improved RetinaNet

Visual ship image object detection has essential applications for near-shore ship management and military object location. In recent years, object detection technology based on a deep learning algorithm has been widely studied in object detection of visible ship images, and achieved outstanding results. However, due to the difference and overlap of nearshore ship objects, the object loss rate is high. Aiming at the above problems, this paper proposes an improved RetinaNet ship object detection algorithm. Firstly, channel attention is added after the residual network, and used to enhance the attention to low-frequency information. Secondly, the cyclical focal loss and the CIOU loss function are used to increase the training times of negative samples in the middle of training, which effectively improves object detection accuracy. The experimental results show that the improved RetinaNet algorithm improves the recognition accuracy of ship objects by 2.5%.