Fine-grained image recognition (FGIR) aims to distinguish visual objects belonging to different subclasses within the same category. Existing methods mainly focus on identifying discriminative regions and extracting the most prominent features. However, this approach leads to scale imbalance between the foreground and background of an image. And it tends to focus on extracting features from salient foreground regions while neglecting valuable information present in the background. To address these two challenges, we propose a weakly supervised foreground-background partitioning and feature fusion framework. Specifically, a foreground-background image partition module is employed to separate the foreground and background regions to resolve the scale imbalance in image. We incorporate a feature similarity calculation module to weigh the foreground and background features. To leverage the background information while capturing discriminative regions, we introduce a selective mask feature module. Comprehensive experiments on four popular and competitive datasets demonstrated the superiority of the proposed method in comparison with the state-of-the-art methods.