Research On Citrus Grading And Grasping Based On Deep Convolutional Neural Network And 3D Vision

Citrus reticulata Blanco,the number one fruit planted in the world,has continued to refresh the total yield.However,the huge citrus yield and backward citrus production mode lead to the problems of low quality and low efficiency in citrus classification and sorting.In this paper,the method based on deep learning is adopted to achieve the accurate classification of citrus,and a 3D visual disorder grasping platform is built to complete the efficient sorting of citrus.The main research results are as follows:(1)To solve the problem of fast classification of citrus with or without defects,an attention mechanism was added on the basis of Res Net34 to improve the weight of useful information,reduce the weight of irrelevant information,improve the collection level of characteristic information of classification module,and enhance the classification ability of the model.The experimental results show that the improved residual network based on attention mechanism can detect citrus classification with99.02 % accuracy.(2)Based on the YOLOv3 network,Mosaic data enhancement,Complete Intersection Over Union(CIOU)loss function and spatial pyramid pooling module(SPP)were added to solve the classification problem of citrus fruit with no surface defects,so as to accelerate the convergence ability of the model and realize the extraction of multi-scale features.Experimental results show that the average accuracy of citrus grading based on Yolov3-SPP network reaches 95.08 %,which is 2.52 percentage points higher than that of YOLOv3,and the overall detection performance is improved to a certain extent.(3)Aiming at the sorting problem of already classified citrus fruits,the method based on deep learning combined with 3D visual guidance disordered grasping robot is adopted to sort the randomly distributed citrus fruits.The experimental results of citrus grading and grasping system show that the system can realize the precise location and grasping of citrus fruits under the condition of high degree of scatter.The classification and classification method of citrus fruits based on deep learning has high accuracy and stability,and provides corresponding technical support for classification and classification recognition of citrus fruits.Through the effective combination of deep learning and 3D visually-guided disordered grasping robot system,citrus fruit grasping task can be efficiently completed in disordered environment with a high success rate.In the work of fruit grading and sorting,the labor consumption is reduced,the quality of citrus grading and sorting is improved,and the productivity of citrus industry is improved.