Discrimination Based on Image Processing at Apple Harvest (Part 1)

Binarization of Fruit Pictures Taken from Apple Orchards

Abstract

This study intends to develop a vision sensor detecting system necessary for robotics of apple harvesting. In this paper, an image processing algorithm using color was investigated as a target such that the rate of fruit pixels segmented correctly (RFPSC) was obtainable above 80% in binary images. The binarization was treated with the threshold value which was selected by analyzing the individual histogram of fruits and leaves and others in color TV pictures taken from apple orchards during harvesting period. The results showed that the RFPSC above target of 80% was obtained by applying -5 in threshold value of chrominance G-Y for red apples in front light, but it was difficult to obtain the level in back light because of low brightness of apples. The RFPSC for yellow-green apples was over 80% by applying 30 in threshold value of difference between two original colors R-B when the brightness of apples exceeded 110, but it was not usually easy to reach the level in back light and discriminate fruits from leaves according to the reflection of sunshine onto leaves.