In view of the difficulty of direct manual maintenance and the complexity of internal environment of GIS compartment

a kind of a machine vision foreign matter recognition algorithm

which is suitable for the special environment of GIS compartment

is proposed in this paper. Since the size of the foreign matter in the compartment is different

the improved multi-scale FT algorithm is adopted in this paper. The Gini index is used to select the optimal significant graph. The way of adaptive threshold segmentation is used to segment foreign matters and the small noises is eliminated by image opening operation. Finally

the improved classification network is used to identify foreign matter. This algorithm

compared with manual detection

has higher accuracy in the field of recognition of small foreign matters and more advantages in detection speed. The comprehensive detection accuracy is not less than 94% and the speed is not less than 50 fps.