Multi-level Attention-enhanced Learning for Fine-Grained Visual Classification

Due to significant intra-class variations and subtle inter-class differences, fine-grained images often pose challenges.In this paper, we propose a multi-level attention-enhanced network framework to address this issue.First, we employ convolutional features to crop the original images to focus on the main part and mitigate the interference of backgrounds on classification.Second, we design a scale-adaptive local discriminative region generation module based on the main part's images, effectively selecting locally detailed and rich image patches.It is worth noting that our proposed method only utilizes image-level labels.Finally, we extensively tested our method on three benchmark datasets, including the CUB-200-2011, Stanford Cars, and FGVC Aircraft.The experimental results demonstrate that our method achieves comparable or better classification performance than previous state-of-the-art fine-grained visual classification methods.