At present, the main algorithm used in feature extraction of apple is the traditional Hough algorithm. The algorithm has the disadvantages of complicated operation and poor real-time performance. In view of this situation, a new algorithm of apple fruit feature extraction is proposed in this paper. A sliding Gaussian template and apple image are convoluted to extract apple circles in this algorithm. Experiments and simulation results show that this method can achieve high accuracy of apple fruit detection under single, adjacent and overlapping conditions. In the case of a single apple, the detec-tion accuracy of apple fruit can reach 96.6%. Even in the case of contiguous and overlapping, its recognition accuracy can reach 94.1%. It fully meets the needs of apple's real-time, efficient grading.