We propose an estimation method for aircraft target direction based on histograms of oriented gradients (HOG), and then use the improved active shape model (ASM) to model the deformation between the different types of targets. Finally, we use kernel density estimation method (KDE) global statistical shape constraint to obtain the target to achieve the target recognition, and design a semi-automatic image feature point detection strategy for aircrafts, which improves the efficiency of training samples for calibration feature points. Recognition experiments on aircraft remote sensing images show the proposed method can better recognize aircraft targets than the existing methods.