Abstract: In order to solve the problem of grasping failure or missing grasping due to the change of target gangue position and posture caused by belt slip, deviation and belt speed fluctuation of belt conveyor when the gangue sorting robot sorts gangue, an improved ORB-FLANN efficient matching method of gangue recognition image and sorting image is proposed. An improved ORB feature point detection method is proposed to detect the feature points in the recognition image and sorting image of coal gangue, so as to realize fast detection of image feature points; An improved FLANN matching algorithm is proposed to match the image feature points to achieve efficient matching between the recognition image of coal gangue and the sorting image. Aiming at the problem of long time and low repetition rate of traditional ORB method for coal gangue image feature detection, an improved ORB feature detection method is proposed to improve the speed and repetition rate of image feature point detection; Aiming at the low accuracy of traditional FLANN matching method for coal gangue image matching, an improved FLANN matching method integrating PROSAC algorithm is proposed to eliminate the wrong feature matching point pairs and improve the accuracy of image matching. The method, SURF feature matching method, HU moment invariant matching method, SIFT feature matching method and ORB feature matching method are applied on the experimental platform of the double mechanical arm truss type gangue sorting robot independently developed by the team to carry out gangue matching experiments under different belt speeds, scales and rotation angles. The results show that the matching rate of the method in this paper is 98.2%, and the matching time is 141 ms. It has the characteristics of high matching rate, good real-time performance and strong robustness, It can meet the requirements of efficient and accurate matching of gangue recognition image and sorting image.