MTI-YOLO: A Light-Weight and Real-Time Deep Neural Network for Insulator Detection in Complex Aerial Images

Insulator detection is an essential task for the safety and reliable operation of intelligent grids. Owing to insulator images including various background interferences, most traditional image-processing methods cannot achieve good performance. Some You Only Look Once (YOLO) networks are employed meet requirements actual applications detection. To a trade-off among accuracy, running time, memory storage, this work proposes modified YOLO-tiny (MTI-YOLO) network in complex aerial images. First all, composite collected common scenes “CCIN_detection” (Chinese Composite INsulator) dataset constructed. Secondly, improve accuracy different sizes insulator, multi-scale feature headers, structure fusion, spatial pyramid pooling (SPP) model adopted MTI-YOLO network. Finally, proposed compared trained tested on dataset. The average precision (AP) our 17% 9% higher than YOLO-v2. Compared with YOLO-v2, time slightly higher. Furthermore, usage 25.6% 38.9% lower YOLO-v2 YOLO-v3, respectively. Experimental results analysis validate that achieves performance both backgrounds bright illumination conditions.