A novel image processing algorithm to separate linearly clustered kiwifruits. (July 2019)

• Record Type: Journal Article
• Authors: Fu, Longsheng; Tola, Elkamil; Al-Mallahi, Ahmad; Li, Rui; Cui, Yongjie
• Is Part Of: Biosystems engineering. Volume 183(2019)
• Page Start: 184
• Page End: 195
• Publication Date: 2019-07
• Subjects: Machine vision -- Segmentation -- Detection -- Calyx -- Counting; Bioengineering -- Periodicals; Agricultural engineering -- Periodicals; Biological systems -- Periodicals; Génie rural -- Périodiques; Systèmes biologiques -- Périodiques; 631
• DOI: 10.1016/j.biosystemseng.2019.04.024
• Languages: English
• ISSNs: 1537-5110

Abstract

Abstract : This research work aims at developing a machine vision system capable of distinguishing kiwifruits on plants prior to harvest. The methodology was based on developing an algorithm able to detect each fruit, even when they are clustered in a line. It segments the fruits from the background, counts the number of fruits in each cluster, and identifies the edges of each fruit. After segmentation, the algorithm initially distinguishes between the fruit skin and calyx based on colour differences using selected hue and red channels. Next, a calyx line is drawn to connect all the calyxes in one cluster together. Then, the periphery of each cluster is scanned to find the contact points between the adjacent fruits. Finally, a separating line is drawn between the two closest contact points, provided that this line intersected almost vertically the calyx line. The separating lines determine the borders of each fruit and enable singling them out. The results showed that 93.7% of the fruit calyxes were correctly detected. In night-time with flash, 92.0% of the fruits were separated and counted correctly by the algorithm. Highlights: An algorithm detects kiwifruits one by one when they are clustered on a line. The algorithm works day and night in counting kiwifruits and finding boundaries. 2-fruit cluster counting and boundary finding accuracy 97% and 91% respectively. The true positive detection rate of fruit calyxes was 93.7%. Accuracy did not drop below 90% for up to 4-fruit clusters.