Author(s)

Junjie Chen, Siqi Pan, Yanping Chan, Yuedong Ni, Donghua Ye

Corresponding Author

Junjie Chen

ABSTRACT

Artificial intelligence has made lots of achievements in the field of substation condition monitoring. However, in real world, due to the serious environment and small volume of data, the monitoring accuracy and the stability are poor. Therefore, a new method based on pruning YOLOv5 and spatial multiscale data augmentation (PYSMDA) is proposed for the substation condition monitoring. Firstly, an improved multiscale data augmentation method is proposed. According to different distribution, the spatial multiscale convolution is generated and used to enhance the defect image so that it increases the scale of data and the diversity of image data. The improved spatial multiscale data augmentation method weakens the interference of the varying environment on the recognition accuracy, and improves the defect detection accuracy. Then, the YOLOv5 is applied to train the multi-scale image data. For reducing the effect of the YOLOv5 large-scale parameters on the stability of the model, the model pruning method is utilized to shrink the structure parameters to improve the defect identification accuracy. The effectiveness of the proposed method is evaluated on the substation defect images. Experimental results indicate that the proposed method is well-performance for substation condition monitoring.

KEYWORDS

Substation condition monitoring, Yolov5, Multiscale data augmentation, Model pruning