Research On 3D Visual Classification And Recognition Algorithm In Fruit Sorting Robot

With the rapid development of computer vision and robotics,vision-based intelligent robots have been widely used in military and civilian applications.At present,fruit sorting and retail industries require fruit sorting robots.In this topic,research on key technologies of fruit sorting robots based on 3D vision is carried out according to market demand.Aiming at the 3D structure and surface characteristics of fruits,the main research is based on the extraction and classification of fruit features.The specific content is as follows:First,pre-process the point cloud data,that is,outlier filtering to eliminate noise points and Voxel Grid-based downsampling to simplify the number of point clouds and reduce the amount of calculation;then,extract the local structure topology and surface convexity and concaveness of the preprocessed 3D model,Color,texture and other structural and surface features,and explore the importance of the features according to the PCA projection direction to construct feature vectors to reduce the impact of background and local occlusion on feature extraction.It is verified by experiments that the algorithm in this paper can efficiently extract the features of point cloud data.Aiming at the problem of selecting local and global features of point cloud data in the current object classification and recognition algorithm,this paper uses the improved DGCNN algorithm as the object classification algorithm framework.Firstly,the distance weight and KNNG are adaptively integrated to enhance the effective information of the sampling points and improve the local neighborhood topology.Second,based on the extracted 3D features,the network is refined based on the initial network framework of DGCNN to improve the ability to express object features.Finally,the overall network framework is established based on the logic of Point Net to complete the segmentation and classification of point cloud.After experimental verification,the algorithm's recognition classification rate is 0.1% and 0.6%higher than the DGCNN algorithm,respectively.In the algorithm verification stage,feature extraction and classification recognition are combined to achieve the task of sorting the targets in the scene on the sorting robot.