UAV Maritime Target Detection Algorithm Based on Improved YOLOv7

• Main Authors: Feng, Haoyu, Ge, Yuan, Ye, Gang, Wang, Jingpin, Zhang, Lufei, Zhao, Jiqiang
• Format: Conference Proceeding
• Language: English
• Subjects: Accuracy; Autonomous aerial vehicles; Convolution; Coordinate Convolution; Feature extraction; Maritime Target Detection; NWD Loss Function; Sea surface; Shape; YOLO; YOLOv7
• ISSN: 1948-9447
• DOI: 10.1109/CCDC62350.2024.10588282

On sunny days, due to the different flight heights and shooting angles of the UAV on the sea, there may be solar flares projected on the sea surface and mutual blocking targets in the aerial image, which may seriously interfere with the detection process. To solve the above problems, this paper proposes an unmanned aerial vehicle(UAV) maritime target detection algorithm based on improved YOLOv7. First, the YOLOv7 header network is improved by adding Coordinate Convolution modules and replacing Repconvolutions with Coordinate Convolutions to enhance the spatial awareness of the network. Secondly, the Convolution Block Attention Module(CBAM) is added into the feature extraction framework of YOLOv7 to improve the detection accuracy of small targets by enhancing the ability to extract the feature information of small targets. Finally, the NWD Loss Function is used to optimize the similarity of overlapping small targets. In this paper, the SeaDronesee dataset is used for testing. The experimental results show that the improved model can effectively improve the detection results of small targets, and the mAP value is 93.4 percent, which is 6.7 percent higher than that of the baseline method, and the effect is better than the mainstream algorithms such as SSD and YOLO series.