With the rapid development of remote sensing technology, research on land use classification methods based on hyperspectral remote sensing images has attracted widespread attention. Existing land-use classification studies mostly use the average filtering method at a single scale for spectral image processing. These methods cannot accurately filter the window range, which leads to the neglect of image detail information, and the single kernel matrix cannot characterize multifeature information, resulting in reduced classification accuracy. Therefore, this study intended to use a superpixel segmentation method to perform multiscale superpixel segmentation on the first principal component of a hyperspectral image at multiple scales. By combining the weighted multiscale spatial-spectral kernel and the original spatial-spectral kernel to form a synthetic kernel for land use classification, the hyperspectral image of the National Mall in Washington DC was used as experimental data to test and analyze this method. The experimental results showed that the classification accuracy of this method on the experimental test set was 98.53%, which is compared with the classification results obtained by the single-scale spatial spectral synthetic nuclear method, the original spatial spectral synthetic nuclear method and the wavelength segmented synthetic nuclear method, the effective classification accuracy with this method was increased by 7.56%. The results prove that this method can effectively solve the problems of the lack of adaptability of the image spectrum and the lack of comprehensive spectral information and can significantly improve the accuracy of land use classification.