A New Strategy for Bird Detection in Coastal Wetlands Based on Improved YOLOv6 3.0

The coastal bird resources in china are very rich, and the wetlands along the coast of Jiangsu have numerous bird resources such as grus
$j$
aponensis and ciconia boyciana, making it a world natural heritage site. In order to better conduct research on bird detection in coastal wetlands, this paper proposes a new bird detection strategy based on improved YOLOv6 3.0. Firstly, we collect and filter specific bird images from coastal wetlands, then annotate the images and construct the bird image dataset. Further we embed the SEN et attention mechanism into the backbone network of YOLOv6 3.0. Finally, pre-train the YOLOv6 3.0 model by using the public dataset CUB-200-2011, then transfer the obtained model weights to the improved YOLOv6 3.0 model. The experimental results show that the improved model achieves a detection accuracy of 98.40% and an AP value of 98.43% on the coastal wetland grus japonensis dataset. The detection precision and the AP value of the improved YOLOv6 3.0 model increase by 1.30 percentage points and 1.28 percentage points respectively with the original YOLOv6 3.0 model at almost the same speed. The experimental results show that the proposed new model and method can effectively improve the detection performance of grus japonensis.