<p>Meta-learning-based few-shot object detectors often face challenges like feature semantic bias and class distribution bias due to limited sample availability. These challenges stem from incomplete feature semantic representations and imbalanced class distributions. Spatial–frequency-domain fusion analysis, a powerful technique for extracting global features and enhancing key frequency-domain information, provides a new approach to address these issues. In this context, this paper introduces a spatial-frequency consistency and bias-corrected for few-shot object detection. To address feature semantic bias, we propose a multi-frequency feature fusion module that decouples image-level features into non-uniform low- and high-frequency components, which are then fused with multi-scale features to improve both local and global semantic understanding by leveraging semantic information at various frequency levels. For class distribution bias, we introduce a space-frequency distribution consistency module that jointly models structural information in the spatial domain with global properties in the frequency domain. By aligning the reconstruction process in the spatial domain with frequency-domain distribution consistency, our model becomes adaptable to diverse data distributions. Additionally, we propose a frequency-domain distribution alignment loss function to mitigate class imbalance by aligning class distributions across frequency layers. Experimental results in the PASCAL VOC and MS COCO datasets demonstrate that our approach outperforms traditional methods and baseline models such as VFA, showing significant improvements in detection accuracy, stability, and adaptability.</p>