Fine-grained image classification is to recognize hundreds of subcategories belonging to the same basic-level category. Existing methods focus on discriminative localization. However, they have two limitations: time consumption and labor consumption. Therefore, we propose a weakly supervised discriminative localization to address them simultaneously. End-to-end network is to localize discriminative regions and code discriminative features at the same time. Saliency-guided localization learning utilizes attention to avoid using human annotations. Compared with state-of-the-art methods, our propose approach achieves the best classification accuracy and efficiency.