Abstract: Gray level difference index is proposed to quantize the gray level differences between two pixels in a multispectral image theoretically.In the view of Top-of-the-Atmosphere Effect(TOA),the weak information in multispectral image can be suppressed in the strong reflective/rediation field.A novel algorithm is developed to detect edge feature homogeneously and weaken or suppress the effects of TOA.In spectral vector space,the gradient magnitude and direction of a multispectral image is charectrized by the eigenvalues and eigenvectors by the first fundamental form.And dyadic B-spline wavelet transform is applied to obtain edge information in scales from fine to coarse.The results show that the response of gradient is more homogeneous and significant than Canny and Sobel detector results,and integration multiscale edge feature can locate the edge points accurately and ignored the nosie affection.