When segmenting the pixel area of typical foreign matter,it is difficult to select the best segmen-tation threshold,resulting in poor positioning accuracy of the identified typical foreign matter.Therefore,an infrared image recognition algorithm of typical foreign matter for basin insulator flashover is proposed.The impulse noise and Gaussian noise of infrared image are removed,and through edge sharpening and gray homogenization,the image target and background can be divided.Searching the gray variance and maximum value of various images by genetic algorithm and taking the segmentation threshold corresponding to the maxi-mum value as the best threshold,and the typical foreign object pixel areas are segmented to construct an im-proved BP neural network.The foreign object category characteristics of the segmented image are extract and the surrounding typical foreign object categories are identified after iterative training.The experiment results show that this method reduces the distance between the positioning boundary frame and the foreign object rec-tangular frame,and improves the foreign object target positioning accuracy and recognition efficiency.
维普
万方数据