ABSTRACT
Ship visible images detection based on computer vision plays an important role in the field of intelligent ship. To increase the model speed, accuracy, and reduce the parameters of the model to facilitate the deployment on hardware devices in practical applications, this study proposed a new model named STD-Yolov5. Firstly, the attention mechanism module of ECA was embedded in backbone to enhance the feature extraction capability of the network. Secondly, GAFPN was designed to reduce the parameters and GFLOPs. Thirdly, to solve the problem of ship-type false detection and missing detection, this paper presented a new receptive field amplification module named GSPP. Finally, replaced the GIoU bounding box regression loss function with a simpler generalisation of α-GIoU to improve the accuracy of the model. Compared to Yolov5, the [email protected]:.95 of STD-Yolov5 increased by 1.2%, the parameters decreased by 24.85%, and the GFLOPs decreased by 14.46%.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Data availability statement
Data are openly available in a public repository. The data that support the findings of this study are openly available at http://www.lmars.whu.edu.cn/prof_web/shaozhenfeng/datasets/SeaShips%287000%29zip.