Recognition of Mature Citrus in Natural Scene under the Occlusion Condition

Abstract

The recognition of mature citrus in natural scene under the occlusion condition has become research difficult problem of the intelligent robot and its vision because the agricultural environment is diversiform and complex. First of all, the quantitative concept of occlusion and its classification method were proposed. The occlusion rate of the citrus fruit was defined and the occlusion was divided into noneocclusion, mild occlusion, moderate occlusion and severe occlusion, the four cases based on the occlusion rate. Secondly, the Cr component image of the YCbCr color model of the citrus image was segmented by using the improved Fuzzy C-means clustering threshold (FCM) segmentation method, then morphology operation and connected region label were used to remove random noise. Finally, an optimized Circular Hough transform algorithm was proposed to circle fit and extract centroid coordinates and radius, and then the fruit shape was recovered. A mature citrus recognition software system in natural scene under the occlusion condition was independently developed based on the Open CV function library. The operational experimental results showed that the recognition rates of our algorithm were respectively up to 98.5%, 97.3%, 94.3% and 55.6% for the cases of the none-occlusion, mild occlusion, moderate occlusion and severe occlusion, which are higher than the previous and could meet the requirements of fruit picking robot.