Recognition of green apples based on fuzzy set theory and manifold ranking algorithm

Accurate recognition of green apple targets is one of the most practical technologies applied to agricultural sciences. The primary goal of this study was to solve the difficulties of distinguishing green apples in similar background areas, such as leaves, by fusing fuzzy set theory and the manifold ranking algorithm (FSMR). First, the original images were roughly enhanced to make apple targets more prominent by using fuzzy set theory in the HSI color space. Second, the specific query nodes were selected for ranking pixels in the whole image to obtain the final recognition results. Then, a series of operations including hole filling, edge smoothing and the mathematical morphology method were carried out to extract the contours of apple targets. Finally, the proposed method was compared with the additional methods based on manifold ranking (MR), saliency detection by self-resemblance (SDSR) and the K-means clustering algorithm. 100 images of green apple targets with different statuses in natural scenes were tested in the experiment. The results showed that the proposed method could effectively extract the real contours and achieve a high degree of coincidence with the edge lines of the targets, indicating that the recognition results were more accurate. Additionally, the average overlapping coefficient was 90.87%, which was 17.34% higher than the SDSR algorithm and 15.98% higher than the K-means clustering algorithm. In conclusion, the proposed method could provide a reference for the developments of growth monitoring, yield estimating and automatic fruit picking.