Ship detection technology plays a major role in maritime traffic safety and waterway management. Current ship detection algorithms mainly rely on radar remote sensing images, which cannot achieve accurate and real-time ship detection. In order to solve this problem, this paper proposes a ship detection method based on an improved YOLO algorithm. Firstly, we introduce the ASPP module to enhance the model's ability to distinguish foreground from background by fusing detailed information of spatial features at different scales and reinforcing semantic information. Secondly, we introduce the CBAM attention mechanism, which makes the neural network pay more attention to the location and spatial information and more attention to the region of interest, thus reducing the attention to invalid features; through adaptive feature fusion, the network obtains features with richer layers, which improves the accuracy of the network in detecting ships. Finally, Focal Loss is used as the loss function to improve the detection accuracy of hard-to-identify targets and to alleviate the problem of unbalanced datasets. Our experiments on the ship dataset show that our improved YOLO algorithm outperforms other conventional methods in the ship detection task. In addition, we also performed a practical application of our method and found that it achieves an average accuracy mean of 96. 8% in real-time ship detection with high accuracy and reliability.
Ship detection based on improved YOLO algorithm
06.01.2023
1159796 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Small Target Detection Algorithm in Remote Sensing Image Based on Improved Yolo
Springer Verlag | 2021
|