<p>In power systems, substation equipment defect detection is crucial to ensuring the safe and stable operation of the power grid. However, existing object detection methods overlook the differences in channel representations of object regions during feature fusion, leading to insufficient representation of defect features and causing missed and false detections. Moreover, these methods lack effective perception of diverse defect geometries during feature extraction in Backbone, resulting in reduced defect recognition capability. To address these issues, we propose a Region-Focused Spatial-Frequency Domain Fusion and Axial Geometric Spatial Perception Enhancement Network (RFAG-Net). Specifically, we propose a Region-Focused Spatial-Frequency Domain Fusion (RFSFF) module that partitions channels across different receptive fields to focus on regions of varying scales, while exploring the complementary relationship between spatial and frequency domains to enhance the feature representation of defect objects and improve detection performance. Furthermore, an Axial Geometric Spatial Perception Enhancement (AGSPE) module is designed to model the geometric relationships between adjacent features in both horizontal and vertical directions, enhancing the perception of geometric features in target regions. Experiments on the substation equipment defect image dataset demonstrate that our RFAG-Net outperforms several state-of-the-art detection methods.</p>