Abstract: An FDA-YOLOv8n photovoltaic module defect-detection model based on YOLOv8n was investigated to address the problems of imprecise positioning and low recognition accuracy. These problems are caused by small target defects with dense or overlapping locations in the infrared images obtained by photovoltaic modules under the complex scenario of large-scale photovoltaic power plant operation and power generation. First, the FasterNet Block was integrated into the C2f module in the backbone layer to reduce the parameter count and amount of model computation. Then, a deformable attention transformer (DAT) was added after the SPPF layer and used with dynamic sampling points to optimize the extraction of feature maps. This allowed better adaptation to the detection of location-intensive defects. Second, the neck layer was fused with ASF-YOLO, and a P2 small-object detection layer was added to improve the small-object defect-detection accuracy of the model. Finally, a loss function inter-MPDIoU was introduced to enhance the small-object defect-localization ability of the model. Experimental results showed that the proposed FDA-YOLOv8n detection model only used 2.4 M parameters, which was 20% lower than those used by the original YOLOv8n model. Meanwhile, mAP@0.5 improved by 3.5%, and mAP@0.5:0.95 improved by 2.5% compared with the results of the original YOLOv8n model. In addition, the detection accuracies of hotspot and battery string fault defects were improved by 5.1% and 5.5%, respectively, demonstrating the accuracy of the model for small-object defect detection.