Research on Innovative Methods and Implementation Paths of Ship Target Detection in Deep Learning

• Published in: Computer fraud & security pp. 13 - 33
• Main Author: Kaizhi Li
• Format: Journal Article
• Language: English
• Published: 25.12.2024
• ISSN: 1361-3723; 1873-7056
• DOI: 10.52710/cfs.86

In order to improve the recognition accuracy of ship targets in SAR images, this paper adopts a data-driven deep learning ship detection scheme, and attempts to use YOLOv5, YOLOv7, and YOLOv7-X to recognize ship targets in the HRSID dataset and a self-constructed HRSID dataset enhanced for small targets, and finally presents them in a visual form. While recognizing images, it is also possible to recognize ship targets in videos and real-time cameras. This article mainly explores in depth from four aspects: selection and reinforcement of datasets, selection of deep learning algorithms, Python scripts and visual interfaces for assisting model training. This article aims to simplify the repetitive operations in the experimental process and further enhance the convenience of detection. Relevant Python scripts and visual interfaces have been designed and written to achieve auxiliary functions such as batch and random addition of small targets in images, stacking of real boxes and recognition boxes of test samples, and improving the efficiency of ship detection. The experimental results in the article show that using YOLOv7-X and the enhanced HRSID dataset can achieve a recognition accuracy of 92.53%, and also have good recognition effects on small targets and ship targets in complex backgrounds such as ports. Compared with traditional ship detection methods, there is a significant improvement in recognition accuracy.