Aiming at the problem that the object detection technology based on deep learning cannot be applied in the industrial field to realize efficient and accurate detection on mobile embedded devices, a lightweight feature fusion detection model (LFF-DM) for real-time detection of transmission line components based on improved YOLOv3 is proposed in this paper. On the one hand, the clustering results are obtained by the improved K-means algorithm; on the other hand, a lightweight network structure is designed by combining depth separable convolution and inverted residuals. Experiments are carried out on the self-built professional inspection data set including insulator, overhanging wire clip, anti-vibration hammer, birds nest, and ground guidewire, and the results show that the detection speed of 25 FPS and the detection accuracy of 90.48%mAP can be achieved on NVIDIA Jetson AGX Xavier equipment, which is suitable for real-time accurate inspection of transmission line mobile terminal.