A Giant Panda Face Alignment Method Based on an Improved FAN Network

Alignment of giant panda face is a key step in individual identification of giant panda faces. Effectively aligning key points on the giant panda face can help accurately capture information across the entire giant panda facial region, which is conducive to better individual recognition of giant panda faces. Aiming at the problem of lack of effective panda face alignment methods, this paper proposes a panda face alignment method. The alignment method is based on an improved FAN face alignment network model, with the convolutional layer structure of the residual module modified and combined with the channel attention module to make it more suitable for the Panda face alignment task. The experimental results show that the improved alignment network model achieves a Mean Normalized Error (MNE) loss of 4.69%, a failure rate of 27.7% at a threshold of 5%, and a failure rate of 1.80% at a threshold of 8%, which is better than the original FAN network. At the same time, it can align not only relatively straight images of giant pandas, but also images with large pose changes or occlusions, and it has good robustness.