Yolov5s-MSD: a multi-scale ship detector for visible video image

• Published in: Multimedia systems Vol. 30; no. 1
• Main Authors: Chen, Yan-Tong, Zhang, Yan-Yan, Wang, Jia-Liang, Liu, Yang
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.02.2024; Springer Nature B.V
• Subjects: Algorithms; Computer Communication Networks; Computer Graphics; Computer Science; Cryptology; Data Storage Representation; Multimedia Information Systems; Operating Systems; Performance evaluation; Real time; Scale invariance; Special Issue Paper; Visible video ship detection; Attention module; Reparameterized convolution; Multi-scale ship
• ISSN: 0942-4962; 1432-1882
• DOI: 10.1007/s00530-023-01196-6

As the ship multi-scale phenomenon is very common in visible video, it is an important factor affecting the performance of visible video ship detection. Based on YOLOv5s, this paper proposes a real-time multi-scale ship detection algorithm. First, reparameterized convolution is adopted to increase the network width, thereby enhancing the network’s ability to express multi-scale ship features. Second, the depth of SPPF (Spatial Pyramid Pooling Fast) is adjusted to enhance the scale invariance of ship features extracted from the network. Third, the attention module is combined with feature pyramid network to enhance the network’s ability to focus on multi-scale ship features. Finally, confidence propagation cluster is used for post-processing to make the network generation more confident and closer to the boundary box of the real box. The experiment shows that our method can achieve state-of-the-art visible video ship detection performance on multiple evaluation indicators, such as mAP-IOU@0.5, mAP-IOU@[0.5:0.95], APS, APM, APL and so on. And it can meet the requirements of real-time detection.