Research On Visual Algorithms For Automatic Sorting Of Factory Items

The automatic sorting task is complex because of the variety of factory items,different materials,different colors and the spatial location relationship with the background.The automatic sorting device of factory goods mainly includes visual algorithm and dynamic device.This paper mainly studies the visual algorithm part.With the rapid development of in-depth learning and computer vision technology,enterprises and universities at home and abroad have attached great importance to automatic sorting technology based on machine vision,and have been greatly applied in factory production.Deep learning can effectively select and represent features,which brings new development opportunities for computer vision,robotics,finance and medical treatment.The application of in-depth learning in the field of factory automation production has become the same consensus and challenge,and has also brought new impetus to the acceleration of its work.The main contents of this paper are as follows:Firstly,the history and characteristics of in-depth learning are introduced.By building in-depth models and learning more useful features from a large number of data,it can obtain high accuracy of classification or prediction.The development,structure and properties of convolutional neural network and its application in semantics segmentation and two-dimensional and three-dimensional information fusion are emphatically expounded.Aiming at the problem of automatic sorting of factory goods,this paper analyses the advantages and disadvantages of the mainstream semantic segmentation network and the semantic segmentation network based on the generation of antagonistic network in detail,and finally designs the semantic segmentation network for factory goods based on the cyclic generation of antagonistic network.For data information extraction,pairs of data are used to improve more semantics segmentation information,and feature extraction network is added to enhance the feature extraction and learning ability of the network.Finally,the loss function of the network is optimized,so that the final semantics segmentation ability of the network reaches the best level at present.Then,an end-to-end two-dimensional and three-dimensional information fusion algorithm is designed.Firstly,the Two-Dimensional Semantics segmentation algorithm is used to extract the two-dimensional information.Then,the three-dimensional point cloud features are extracted by using depth network,and the two-dimensional and three-dimensional features are embedded and fused.Finally,the network outputs the location and spatial information of objects.Finally,the semantics segmentation algorithm and two-dimensional and three-dimensional information fusion algorithm designed in this paper are tested.The performance of the network on the test set has reached a high level,which meets the design expectations.The network designed in this paper has a certain practical significance.