In view of shadows or dark regions of composite insulator surface under the condition of natural light

insulator cannot be accurately segmented. Consequently

we put forward a kind of composite insulator images segmentation method based on improved color differences. Firstly

the area of the composite insulator in the collected image is divided into three parts

namely

the bright

normal

and dark. After the light and dark areas are compensated separately

the insulator image is re-synthesized. Secondly

according to color component extraction and analysis

an improved color difference method is presented to obtain the optimal color difference image. Finally

the Otsu adaptive threshold algorithm combined with the region mark and geometric shape analysis is used to segment insulator

so as to realize the accurate identification of the non-uniform illumination composite insulator of the transmission line. The experimental results show that the proposed method can provide a complete and accurate segmentation of the insulator

which provides a useful reference for the subsequent insulator feature extraction and fault analysis. By comparing the method with the traditional threshold segmentation method and the 2 R-G-B（Red

Green and Blue） color difference method

the average segmentation error is 8.18%

36.12%

27.88%

respectively

and the segmentation error is reduced by 27.94% and 19.70%

especially for low light and strong light

the insulator segmentation efficiency is increased by more than 30%.