In this project, a near infrared spectrometer is used to obtain the spectral images of typical defects of composite insulators of high-voltage transmission lines, and the HOG features and LBP features in the spectral images are extracted based on image processing techniques. The two are added to the support vector machine model (SVM) in a fusion way, and the construction of a typical defect recognition model for composite insulators is completed by iterative training, and the model in this paper is verified and analyzed based on simulation experiments. The LBP-HOG-SVM model in this paper has significant recognition effect compared with the original Faster R-CNN model, in which the recall and precision of this model are 9.76% and 9.2% higher than the original model for glass-type insulators, and 11.49% and 13.06% higher than the original model for composite insulators, which confirms that the LBP-HOG-SVM model has significant recognition effect in the identification of typical defects of composite insulators of ultra-high-voltage transmission line, and that the model has significant recognition effect compared with the original model. Line composite insulators in the field of typical defect identification technology.