Research On Recognition And Location Of Apple Picking Robot Based On Machine Vision

Picking robot is a typical intelligent agricultural equipment,which is widely used in various complex picking environments.Picking robot,as a representative of modern intelligent equipment,integrates advanced technologies such as robots,sensors and controllers.In the past five years,artificial intelligence technology has been continuously sought after by the majority of scientific researchers,including the development of intelligent agricultural technology.At the same time,the technical bottleneck restricting the performance improvement of picking robot mainly lies in visual recognition,image processing and path planning.In this paper,in order to improve the picking efficiency of intelligent picking robot,it is necessary to provide reference for practical application under the condition of specific target growth.The intelligent apple picking robot is taken as the research object,and the visual recognition,target positioning,mechanical arm structure design and software design are studied.The research contents and conclusions of this paper are summarized as follows :Firstly,images of apple fruit and its growth under natural light were collected.According to the actual growth characteristics and picking requirements of apple,an apple picking model under natural environment based on improved SSD neural network was proposed.The recognition ability of apple picking robot for small target and complex environment is improved by adding attention mechanism and feature fusion.And put forward two different improvement methods : SSD-F and SSD-FC,after experiment and the original SSD neural network training comparison.The experimental results show that the SSD-FC method in the improved model not only has the best detection effect for small targets and complex environment areas,but also has no significant influence on the recognition stability due to the increase of model complexity,which achieves the expected effect.Secondly,the target positioning system of the apple picking robot is built,and the calculation and analysis of the transformation relationship between the world coordinate system and the pixel coordinate system are completed.The selection and calibration of the depth camera are carried out.Because the coordinate information of the picking target obtained by the end effector needs to be converted from the camera calibration results,the manipulator design,the manipulator D-H model construction and the forward kinematics analysis will be carried out.Eye-in-hand method was used to build the visual picking system and conduct the hand-eye calibration simulation experiment,and Tsai calibration method was used to obtain the position relationship results of hand-eye coordinates.Finally,the visual recognition and positioning system of apple picking robot was experimentally verified.The Intel Real Sense camera is used to identify and locate apple targets,and the three-dimensional coordinate information is presented in the recognition result box in real time to realize system visualization.The rationality of the improved algorithm and hand-eye calibration conversion results is verified by experiments.The apple intelligent identification software based on PyQt toolkit was designed.