Edge detection is one of most important fields in computer vision,and lots of algorithms have been proposed by former scholars.Based on properties of image and detector,and taking relevances as criterion,this paper proposes an algorithm which use genetic algorithm to select optimal wavelet transform scales to form an adaptive Gaussian wavelet scale space.By fusing image's edges detected at different scales in this scale space,image will be improved in edge details and noise immunity.Experiments show that this adaptive scale wavelet algorithm has advantage in terms of noise removing and corner points positioning over Canny algorithm,single-scale wavelet algorithm,binary scale wavelet algorithm or multi-scale wavelet a algorithm.