Recognition of apple fruit in natural environment

Abstract

The recognition method for apple fruits in natural environment was developed. Firstly, the acquired apple images were executed characteristic analysis and classified by difference of light acceptance and growth state. Secondly, the OTSU dynamic threshold segmentation method with I2 color characteristic in the I1I2I3 color space was chosen through comparing the apple image segmentation methods based on the color feature in different color spaces. Nextly, the image perfection and noise removal were carried out for the above segmentation image. Thirdly, the apple fruit contour model was established. The apple fruits were recognized by edge detection and the improved RHT transformation method, the overlapped apples and severely occluded apples by the branches and leaves were respectively done the separation and restoration operations before they were recognized. Finally, some fruits recognition experiments were done for apples in three different states with the non-occluded, overlapped and severely occluded by branches and leaves. The results showed that this proposed method was feasible and can basically meet the requirements of robot picking.