Target recognition method of green pepper harvesting robot based on manifold ranking

Abstract

In order to solve the problem of the recognition of green pepper harvesting robot in the near color background, a target recognition method based on manifold ranking is proposed in this paper. After pepper image enhancement by the local contrast enhancement algorithm, super-pixels extracted via energy-driven sampling (SEEDS) is used to construct super-pixel blocks of the enhanced image, the upper, lower, left and right boundary query nodes are used to query the boundary of the super pixel block image respectively. Then the manifold ranking is used to sort the image boundaries to obtain four saliency maps, and the final saliency map is obtained by fusion. Finally, the noise is removed by morphological operations and the contour of green pepper target is obtained to realize the recognition of green pepper in complex environment. Compared with classification and regression tree (CART), conditional random fields (CRF) and threshold, the proposed method can effectively identify the green pepper target, with the recognition accuracy of 83.6% and recall rate of 81.2%. The performance index is obviously better than the other three methods, which can meet the requirements of the actual operation of the harvesting robot.