To solve the problem of traditional pollution detection methods on the prevention and control of pollution flashover of transmission line insulators,the non-contact and high-resolution hyperspectral technology is used to study the on-line pollution detection technology. At the same time,an insulator pollution level identification technology based on wavelet packet energy spectrum feature optimization is proposed to effectively extract the spectral features reflecting the pollution degree and weaken the influence of redundancy and interference information. Firstly,the spectral images of insulator samples with different pollution levels are segmented to extract the spectral mean curve of pixels in uniform pollution area. Secondly,the difference of light intensity uniformity and environmental noise of different images are preprocessed,and the differentiability among different pollution levels is improved by logarithmic transformation. Thirdly,the feature extraction of wavelet packet energy spectrum is performed on the preprocessed spectral lines. Finally,a pollution level recognition model based on the proposed features and support vector machines (SVM) is established. The experimental results show that the SVM pollution level recognition model based on wavelet energy spectrum features achieves 99.8%,and it has higher recognition accuracy than full band data or principal component analysis (PCA) feature data does.