RG-DCNet: Recursive Gating Based Algorithm for Fine-grained Image Recognition

Aiming at the shortcomings of traditional convolutional neural network methods in capturing long-range spatial dependencies in fine-grained image recognition due to the restricted sensory field expansion and local connectivity characteristics, a recursive gated dynamic convolutional algorithm based on convolutional optimization and recursive gated networks is proposed. The algorithm utilizes a dynamic directional convolutional optimization backbone network to combine the features of the input image to automatically enhance the local features of the key parts of the target, and realize cross-layer jumping sense field expansion, so as to enhance the feature contrast between the target and the background. Meanwhile, recursive gated convolution improves the deficiency of traditional convolutional neural network in long-distance dependent modeling, and the recursive structure is used to realize the collaborative extraction of global and local features. The experimental results show that the proposed algorithm achieves 91.7%, 94.77% and 93.85% performance on CUB-200-2011, Stanford-Cars and FGVC-Aircraft datasets, respectively, which proves its improvement in global information modeling and feature extraction capabilities.