Obstacle recognition is one of the key techniques for deicing robot on high voltage transmission line. According to the structure of 220 kV transmission line and the feature of special environment, the images photographed on line by the robot camera can be processed. Images have been binary converted on the basis of threshold optimization after some preprocesses of de-noising, expansion and erosion. Then, image edges were detected by using wavelet modulus maximum algorithm, and the wavelet moments of edge images that have invariance were calculated. The SVM begins to classify images after being put into eigenvectors that have been optimized into the neural network of SVM, which realizes the image classification of obstacle. Simulation experiments have indicated that eigenvectors of wavelet moments are stable and reliable and SVM classifier has a high accuracy of object recognition. It is a feasible method that combines both virtues.