In the aviation field, the detection of surface defects on aircraft skin is crucial for ensuring flight safety. In response to the shortcomings of existing aircraft skin defect detection algorithms in small object detection, this paper proposes an aircraft skin defect detection algorithm based on an improved EfficientDet model. First, the convolutional block attention machine (CBAM) was integrated into the backbone EfficientNet to enhance the model's attention to defect areas. Second, the hierarchical structure and feature fusion strategy of bidirectional feature pyramid network (BiFPN) were optimized and adjusted to further enhance the ability of feature extraction and multi-scale feature fusion for small target defects. Finally, a scale aware loss function was adopted to enhance the robustness of the model in defect detection at different scales. The experimental results on the self built aircraft skin defect image dataset show that the improved algorithm achieves an average detection accuracy of 91.32%, which is 3.69 percentage points and 2.05 percentage points higher than EfficientDet-D0 and YOLOv5s, respectively. It has significantly improved the detection accuracy and performance for aircraft skin defect types such as paint peeling, scratches, and dents.