YOLOv7-sea: Object Detection of Maritime UAV Images based on Improved YOLOv7

Object detection algorithms play an important role in maritime search and rescue missions, where they are designed to detect people, boats and other objects in open water. However, the SeaDronesee dataset has the characteristics of small targets and large sea surface interference, which brings great challenges to general object detectors. To address these issues, we propose an improved detector YOLOv7-sea. Based on YOLOv72, we add a prediction head to detect tiny-scale people or objects. Besides, we integrate Simple, Parameter-Free Attention Module (SimAM) to find attention regions in the scene. To achieve further improvements to our proposed YOLOv7-sea, we provide some useful strategies such as data augmentation, Test time augmentation (TTA), and bundled box fusion (WBF). On the ODv2 challenge dataset, the AP result of YOLOv7-sea is 59.00%, which is about 7% higher than the baseline model (YOLOv7).