Hyperspectral images can provide rich spatial and spectral information, providing significant advantages for target recognition. Due to the large number of bands and data in hyperspectral images, the problem of “same object, different spectra” is quite serious in the process of target recognition. In order to solve this problem, this paper proposes two methods: one is a nonlinear mapping method for search recognition, and the other is a recognition method that combines sparse representation and multi-scale analysis for matching recognition. With nonlinear mapping processing, high-dimensional features of small targets can be accurately extracted, achieving precise recognition; The problem of “same object, different spectra” in hyperspectral image target recognition has been effectively solved through sparse representation and multi-scale analysis, improving the accuracy of target recognition. Compared with the classic target recognition methods CEM and MF, the experimental results show that the proposed method can effectively improve the target recognition performance of hyperspectral images. This has promoted the application of hyperspectral images in fields such as land and resources investigation, geological hazard assessment, urban planning, and disaster relief.