In recent years, numerous Few-Shot Fine-Grained Image Classification methods have been proposed, primarily focusing on better fine-grained feature extraction. Among them, the feature mapping reconstruction network (FRN) is a prominent approach to solving this problem. Nevertheless, extensive comparative experiments reveal that traditional FRN only utilizes support features from a single channel dimension to reconstruct query features, while neglecting interactions between different dimensions, which leads to inaccurate reconstruction errors. To mitigate this issue, this paper proposes a cross-reconstruction network (CRN), which effectively helps the model learn the features across different dimensions, enhancing its applicability to the few-shot fine-grained classification problems. Additionally, we introduce a multi-scale feature enhancement (MCFE) module for feature information, which works in concert with the cross-reconstruction network to capture feature information of images more effectively and make features more specific. Extensive experiments on a baseline dataset demonstrate the superiority of our approach compared to other state-of-the-art methods.