Power line insulator defects can lead to serious failures in transmission systems, making aerial image-based inspection a widely used technique. However, detecting insulators and their defects in aerial images remains a challenging task due to complex backgrounds and varying environmental conditions. This paper presents an automatic detection approach for insulators and their defects using aerial imagery, and proposes a novel deep learning framework based on feature-enhanced CenterNet. To improve insulator detection, we first deepen the backbone network to enhance feature extraction under complex scenes. Then, we incorporate the Res2Net module to expand the receptive field and improve multi-scale representation. Finally, a pyramid pooling module is applied to fuse multi-scale and sub-region contextual information, thereby increasing identification accuracy and reducing missed detections. To further enhance defect detection performance, we integrate an efficient channel attention mechanism during feature fusion to better capture defect-related semantic features. Additionally, we introduce a convolutional attention module in the deconvolution upsampling stage to highlight defect features while suppressing noise. Finally, we adopt CIoU Loss for bounding box regression to improve localization accuracy. Experimental results demonstrate that the proposed framework meets the accuracy and robustness requirements for insulator defect detection. Evaluated on an extended public dataset, the method achieves a mean precision of 95.06% for insulators and defects, demonstrating strong performance across diverse inspection scenarios.