Citrus Fruits Harvesting Sequence Planning Method Based on Visual Attention Mechanism : A Novel Cognition Framework for Citrus Picking Robots

Abstract

In order to improve the intelligence of picking robots, a novel cognition framework for citrus picking robots was proposed to realize harvesting sequence planning inspired by human cognitive behaviors. Firstly, the Kinect v2 camera was utilized to collect RGB-D images of citrus fruits, and the YOLOv4 model was used to detect the citrus fruits. Secondly, based on four color features, R, L*, b* and I, a saliency detection algorithm was proposed to establish the mapping relationship between the saliency and the quality of the fruits. Finally, the object saliency and center depth of fruits were set as decision-making indicators to determine the picking priority of each fruit based on system engineering theory. The in-field harvesting experiments show that, compared to the traditional method, the quality of the fruits harvested by the proposed method increased by 5.34%, indicating that the proposed method is able to help picking robots make smarter decisions from the perspective of the maximum benefits. This study provides case reference and technical support for the realization of intelligent picking robots.