In order to estimate the noise levels of hyperspectral images under ground-based imaging conditions accurately, a residual-scaled local standard deviations (RLSD) method after edge elimination was proposed. Firstly, the obtained hyperspectral image was divided into several sub-blocks of appropriate size, and then the edge information of the image was detected by using Canny edge detection operator, and the sub-blocks containing edges were judged and eliminated. The noise estimation of the uniform sub-blocks after the removal of edge sub-blocks was carried out by the method of multiple linear regression and residual error. The total error of noise was 1.985×103 and 2.197×103 for different sub-regions of the same land-based hyperspectral images by 4×4 pixel and 8×8 pixel segmentation. The results show that the proposed noise estimation method is robust to the noise evaluation of hyperspectral images under the condition of land-based imaging, which provides a reference for the subsequent processing and application of land-based hyperspectral images.