Research On Apple Identification And Picking Point Location System Based On Deep Learning

With the rapid development of robotics and the gradual increase of agricultural labor costs,the design and production of agricultural production robots have become a hot research topic in the farming industry in order to reduce production costs in agriculture,promote agricultural intelligence,and liberate labor.Still,agricultural robots cannot be quickly and practically applied on a large scale due to the limitations of their vision systems.In order to solve the problem of poor recognition and positioning accuracy of the visual recognition and positioning system of apple harvesting robots,this paper investigates the improvement of the recognition efficiency and positioning accuracy of apple targets.A network pooling module is proposed to extract edge features from apple fruit targets in an unstructured environment,and the network model is integrated and trained by constructing a database of apple fruit images in a natural environment.The accuracy and real-time performance of the trained network model for different categories of apple fruit targets are significantly improved compared with the original network.A lightweight network model based on target detection is proposed to locate apple fruits in the natural environment,where it is difficult to determine the spatial information of the target fruit due to the different growth postures of apples.The robustness of the network model is improved.A lightweight network model based on key point detection is proposed to detect and localize apple fruit stalks.To address the problem that apple fruit picking points are difficult to identify and detect in a complex unstructured environment,a hybrid attention module is linked at the end of the feature extraction stage of the network to make the network model pay more attention to the feature information of fruit stalk picking points in order to improve the recognition and detection efficiency of apple fruit picking points.The accuracy of the algorithm model for recognition of different categories of apple fruit targets and the accuracy of target fruit localization and its picking point localization are verified in the laboratory environment and natural environment,respectively,and the application of the algorithm can identify and localize different categories of apple fruit targets,which provides a good technical path for the intelligent application of apple harvesting robot.