Abstract: Currently, the mainstream of line segment detection algorithm mostly utilizes the pixel's gradient orientations feature, but these algorithms exist problems of over-segmentation and being prone to miss long segment and low contrast segment commonly, especially confronting high-resolution color image. In order to solve these problems, this paper proposed a modified extending algorithm based on LSD, the current excellent method. This algorithm makes the best of original high-resolution color image's multichannels and multiscale characteristics. Then it combines robust line segments filtering method based on scale features and multichannels-multiscale merging method. Our algorithm obtains the detection results closer to vision perceptually. Experimental results show that our algorithm substantially descending the miss rate of long segment and low contrast segment, and obviously eliminating the over-segmentation phenomenon.