A deep attention mechanism method for maritime salient ship detection in complex sea background

Abstract:

Saliency ship detection has received increasing attention due to its important applications in maritime field in recent years. Up to now, numerous studies on saliency detection have been done based on traditional methods and deep learning methods. But these previous research works are still not competent enough in detecting ship targets with complex backgrounds and noises. In this letter, we propose a deep attention mechanism method for more accurate and faster maritime salient ship detection. We optimize the initial ship saliency map by using a feature attention module to focus on salient objects. We reduce and improve the convolution kernel in refinement residual module to enhance the detection efficiency. In addition, Leaky ReLU is selected as the activation function to increase the non-linear capability of the method. Experiment results show that, the proposed method could obtain outstanding performance in salient ship detection in complex sea background.