This study, aimed at the problems of spectrum waveform characteristic distinction, operation speed, and spatial detail, proposes an improvement in the algorithm for hyperspectral remote sensing feature recognition. Based on this, we propose a fractal signal algorithm. The performance, efficiency, etc., of the algorithm itself is tested using CASI hyperspectral data and hyperspectral remote sensing image lithologic characteristics of the study area are also extracted. The initial value of the signal, the iteration step length, and other characteristics of the fractal signal in hyperspectral remote sensing data were discarded in this study. To a certain extent, the fractal signal algorithm can refine the distinguishability of similar characteristics in hyperspectral, and when used for feature extraction from CASI lithology data it accurately extracted the surface lithology of exposed bedrock areas.