<p>This study proposes a lightweight fault detection method for small targets to address the issues of slow processing speed and high susceptibility to false and missed detections in fault detection algorithms, particularly in cluttered backgrounds. The proposed method is based on an enhanced YOLOv8 framework that integrates a squeeze-and-excitation network (SENet) with Transformer-based optical character recognition (OCR). The backbone network structure of the YOLOv8 algorithm was optimized by incorporating the channel attention module SENet, thereby enhancing the model performance, adaptability, and generalizability. The differentiable feature learning loss function was refined to better guide the learning process, minimize positioning errors, and improve the overall accuracy. In the neck layer, group shuffle convolution replaced traditional standard convolution to construct a lightweight detection network, which increases the diversity of the receptive fields of the network, captures contextual information more effectively, and reduces the computational complexity. This study innovatively introduced OCR into the YOLOv8 backbone to improve the digit recognition accuracy. A charge-coupled device-type image sensor was employed as the image acquisition device to enhance the fault detection accuracy further. Finally, the experimental results demonstrated that the improved YOLOv8n algorithm achieved a detection precision of 96.84%, a recall rate of 98.73%, and an F1-score of 97.81%.</p>