Visual Recognition Method for Intelligent Picking of Special-Shaped Fruits and Vegetables based on Multimodal Fusion

Abstract

With the acceleration of agricultural intelligentization process, intelligent picking of heteromorphic fruits and vegetables has become a key research direction in the field of agricultural automation. Traditional visual recognition technology has bottlenecks such as low recognition accuracy and weak robustness when dealing with the complex morphology and variable environment of heterogeneous fruits and vegetables. In this paper, we propose a multimodal visual recognition method that fuses visible, depth and near-infrared images, and deeply exploits the features of multi-source data through innovative data fusion strategies and deep learning algorithms. The experimental results show that the accuracy of this method is 92.3% in the recognition of shaped fruits and vegetables in complex scenes, which is more than 20% higher than that of the single-modal method, which effectively solves the core technical problems of intelligent picking of shaped fruits and vegetables, and provides important technical support for the intelligent upgrading of agriculture.