This paper examines the evolution characteristics of the land use structure in karst valley areas on the basis of remote sensing images SPOT with a resolution of 2.5 m and their corresponding socio-economic data in the years of 2004, 2010 and 2015. This study is also conducted by creating a microspatial unit of 200 m×200 m grid to calculate the composite index of the area proportion and land use degree in the inner part of the cell grid, in combination of bivariate local space autocorrelation and hot spot analysis. It is concluded that: (1) The average proportions of forest land, cultivated land, settlements and bare rock in grids in 2004 were 6.46%, 11.76%, 0.87%, and 28.86%, respectively; in 2010, they were 10.36%, 8.77%, 1.27%, and 27.13%, respectively; in 2015, they were 9.13%, 10.30%, 26.18%, and 1.29%, respectively. (2) Various land use types have obvious spatial differences in terms of spatial distribution. (3) Cultivated land hot spot showed a decreasing trend, and later focused on the southwestern part of the study area, the total hot spot area of the forest increased, and the hot spot of bare rock gradually shrank to the northeast of the study area. In recent years, the hot spot area of land use gradually extended to the northwest of the study area, and the Zanthoxylum Bungeanum forest was mainly distributed in the northern dry-hot valley. In short, the area of forest land in the Huajiang valley area increased, and it gradually evolved into concentrated contiguous areas, while the bare rocky land gradually shrank to the north and northeast of the valley area. The diversified development of farmers' livelihoods and the improvement of traffic conditions in the study area reduced land pressure and gradually improved the ecological environment. This paper, taking typical karst geomorphic units as case study areas, reveals the law of land use evolution and its internal differences in karst mountainous areas of Southwest China, which is of great significance to the research on the characteristics and law of land use evolution on a larger scale.