Research On Stacking Target Recognition And Positioning System Based On Deep Learning

Although the traditional target recognition algorithm can quickly complete the target recognition task with a small amount of calculation,it cannot guarantee accuracy in a complex background.Meanwhile,the algorithm performance will not be optimized with the increase of sample data.Therefore,object detection algorithms based on deep learning have begun to be introduced into practical applications.This paper uses binocular cameras and deep learning algorithms to design an automatic identification and positioning system for stacking objects.This system can accurately identify stacking objects and calculate their spatial location.The main innovations of this paper are as follows:(1)Design a system(Binocular-Neural Network Positioning System,B-NPS)based on binocular cameras and deep learning.The system needs to use the "Zhang Zhengyou calibration method" to calibrate the binocular camera and make the "Box" data set to train the neural network ahead of time.B-NPS uses the binocular camera to shoot the space image of the stacking target and input the image into Mask R-CNN to detect the object.Then,calculating the pixel coordinates of the stacking objects in the two pictures.Finally,calculating the coordinates of the stacking target according to the positioning principle of the binocular camera and the internal and external parameters of the camera.(2)With the goal of reducing system complexity and avoiding the waste of computation,the Prejudge Network is added to the system,and the improved system PB-NPS(Prejudged Binocular-Neural Network Positioning System)is proposed.In PB-NPS,the Prejudge Network is added inspired by the "two stage" strategy in the target recognition algorithm,and the picture is judged by the Prejudge Network before subsequent processing.Through the research of Res Net,Inception,and other networks,the Prejudge Network network is proposed based on the Res Net network.The performance of Prejudge Network is tested and verified on the public data set and self-made data set.The results show that the improved PB-NPS system can accurately determine the existence of the target with the help of Prejudge Network,and effectively improve the operating speed of the system.(3)Optimize the convolutional neural network to improve the speed of the PB-NPS system,and propose a new improved system PB-MNPS(Prejudged Binocular-Modified Neural Network Positioning System).After researching the relationship between the number of layers and model performance in convolutional neural networks under general conditions,Mobil Net is used to improve Mask R-CNN to obtain Mobile Mask R-CNN.Mobile Mask R-CNN is added to get an improved PB-MNPS system.Using simulation experiments,testing the performance of the Mobile Mask R-CNN network on the public data set and self-made data set.Results show that the PB-MNPS system runs fast with small memory resources,and has high accuracy in identifying stacking targets.