In order to construct colorful 3-D spatial structure of canopy accurately, apple tree canopies of maturation period were set for research object, and a PMD camera and a color camera were used to acquire multi-source organ images of apple tree canopies, which aimed at studying registration of multi-source images. SIFT algorithm was used to extract characteristic points from multi-source images and RANSAC algorithm optimized by objective function was adopted to purify feature vectors, which overcame scale change and effect of light. Bilinear mapping algorithm was employed as spatial mapping between multi-source images to avoide inaccuracy of using affine transformation to solve space mapping. The registration test in orchards with different natural environment showed that the method proposed was suitable for registration of multi source images of apple tree canopies, and the registration accuracy could be 88.2% in sunny day of light environment, 84.2% in sunny day of backlight environment, 72.7% in cloudy day. The research finding could provide maintenance assurance for picking and yielding links of apple orchard after 3-D reconstruction of canopies.