Few-shot learning for image classification aims at predicting unseen classes with only a few images. Recent works, especially the works on few-shot fine-grained image classification (FSFGIC), have achieved great progress. However, most of them neglected the spatial information and computed the distance between a query image and a support image directly, which may cause vagueness because the dominant objects can exist anywhere on images. A promising solution is to locate salient regions from images for discriminative feature representation learning. This paper develops an automatic salient region selection network without the use of a bounding box or part annotation mechanism for locating salient regions from images. Then a weighted average mechanism is introduced for facilitating a neural network to focus on those salient regions, optimizing the network, and performing the FSFGIC tasks. The experimental results on four benchmark datasets demonstrate the effectiveness of the proposed strategy.