Abstract: Taking the northern part of Yuqiao Reservoir in Jizhou District of Tianjin as the research area, by selecting the panchromatic data of GF-6 and the multi-spectral data of Sentinel-2A. The accuracy of remote sensing identification information of poplar leaf-eating pests was improved by optimizing the three levels of pest sensitive bands, fusion bands and classification recognition features of remote sensing data. After the remote sensing image was preprocessed, the sensitive characteristic bands of pest area identification were selected first, according to the spectral statistical feature difference maximization method. Then the band correlation coefficient was used to achieve the purpose of optimizing the fusion band scheme, and the HPF fusion and GS fusion processing were carried out. Finally, the Relief algorithm was used to optimize 19 features, such as vegetation index, to determine the characteristics of remote sensing identification of leaf-eating pests. The results show that: ① the reflectance values of poplar forests in the red edge band, near-infrared band and short-wave infrared band of Sentinel-2A data were significantly changed through the HPF+GS fusion after optimizing sensitive bands, and the feature contribution was significantly improved during classification. ② After the combination of HPF+GS fusion processing, the details of the fused image were prominent, and the spatial resolution of the red edge sensitive band of the vegetation monitoring spectrum was significantly improved. ③ After multi-level feature optimization, the classification accuracy was significantly improved. The classification accuracy of remote sensing images without optimization was 83.15%, and the Kappa coefficient was 0.6785. The classification accuracy of the image after priority was 90.64%, and the Kappa coefficient was 0.8116. To sum up, multiple feature optimization processing can effectively extract leaf-eating pest image information.