This paper proposes a novel Cross-layer and Cross-sample feature optimization Network (C2-Net) for Few-Shot Fine-Grained Image Classification (FS-FGIC). The C2-Net addresses two critical issues in FS-FGIC: (1) extracting discriminative features while reducing sample-level noise, and (2) achieving feature matching between support and query samples with varying spatial positions and angles. The proposed method consists of two main modules: the Cross-Layer Feature Refinement (CLFR) module and the Cross-Sample Feature Adjustment (CSFA) module. The CLFR module refines features by integrating outputs from multiple layers to suppress sample-level noise. The CSFA module adjusts features through channel activation and position matching operations to enhance shared information and suppress unshared query information. The C2-Net is evaluated on five fine-grained benchmark datasets and outperforms other state-of-the-art methods in most cases. The results show that the C2-Net achieves superior performance in both feature learning and classification, demonstrating its effectiveness in addressing the challenges of FS-FGIC. The code is available at https://github.com/zenith0923/C2-Net.This paper proposes a novel Cross-layer and Cross-sample feature optimization Network (C2-Net) for Few-Shot Fine-Grained Image Classification (FS-FGIC). The C2-Net addresses two critical issues in FS-FGIC: (1) extracting discriminative features while reducing sample-level noise, and (2) achieving feature matching between support and query samples with varying spatial positions and angles. The proposed method consists of two main modules: the Cross-Layer Feature Refinement (CLFR) module and the Cross-Sample Feature Adjustment (CSFA) module. The CLFR module refines features by integrating outputs from multiple layers to suppress sample-level noise. The CSFA module adjusts features through channel activation and position matching operations to enhance shared information and suppress unshared query information. The C2-Net is evaluated on five fine-grained benchmark datasets and outperforms other state-of-the-art methods in most cases. The results show that the C2-Net achieves superior performance in both feature learning and classification, demonstrating its effectiveness in addressing the challenges of FS-FGIC. The code is available at https://github.com/zenith0923/C2-Net.