Animal Image Recognition Method Based on Two-pass Feature Fusion

- The identification and protection of wild animals are crucial to maintaining biodiversity and maintaining ecological balance. Based on the deep learning method and the characteristics of field monitoring images, this paper proposes a Dc_YOLOv5 (Deformable convolution YOLOv5) and NLBP dual-pass fusion network as a feature extraction module. First, due to the randomness of animal activities, the monitoring images may have complex background information. In order to improve the feature expression ability of complex image target detection, a deformable convolutional neural network is introduced in the CSP2_1 module of the YOLOv5 network; in the LBP pathway, the proposed Compared with the traditional LBP method, the NLBP extraction algorithm takes into account the changing relationship between the gray value of the pixel and its neighboring pixels, reducing the loss of feature information; finally, the CAFF-feature fusion module is introduced to reconstruct the feature and Obtain intermediate predictions, effectively focusing on important channels and suppressing irrelevant channels. The experimental results show that this method has excellent recognition effect.