| With the continuous progress of social economy and science and technology,big data and artificial intelligence have also developed rapidly.Computer vision technology,represented by deep learning,is also increasingly widely used in various industries,especially image classification and target detection technology play an important role in promoting technology renewal in medical,transportation,intelligent factory and other fields.The magnetic tile is an important component of the DC motor.At present,the sorting and sealing of the magnetic tile basically depends on the traditional manual classification and testing,which requires a large consumption of manpower and material resources.On the basis of paying attention to the above problems,this paper introduces deep learning into the classification and detection of positive and negative sides of magnetic tiles,which provides a new idea for improving the intelligent level of magnetic tile sorting.The purpose of this paper is to construct a convolution neural network with small model parameters and high accuracy to identify and detect the positive and negative sides of the assembly line magnetic tile in the workshop.The contents of the study are as follows:（1）The lightweight network is used to classify and identify the positive and negative sides of the magnetic tile image.Firstly,the problem of light angle is considered to collect the data set of the front and back of the magnetic tile,and the standard image is obtained by pre-processing.Then Mobile Net v3 network is used to train and classify the front and back of the magnetic tile.The classification accuracy of the system can reach 99.6%.The experimental results show that the method has a small number of parameters,low demand for hardware equipment,has practical feasibility,and has a good effect on magnetic tile classification.（2）The magnetic tile target is detected based on the YOLO v4 network,and the fast and accurate detection of the magnetic tile is realized by lightweight the singlestage target detection algorithm YOLO v4 network to improve the network detection performance.First of all,the backbone feature extraction part of YOLO v4 network is replaced by lightweight network Mobile Net to improve the network detection speed while maintaining a high m AP.Then the enhanced feature extraction part of the modified network is lightweight.The experimental results show that the improved network can improve the detection speed by 40.6% while the detection accuracy is the same as that of the original network,which can achieve real-time detection effect.（3）Because in the actual production,the magnetic tile workpiece may appear on the conveyor belt at any angle,the traditional horizontal frame target detection method will not be conducive to the accurate grasping of the magnetic tile by the robot.In order to solve the above problems,the CenterNet detection framework is introduced into the magnetic tile target detection,and the multi-attitude detection of the magnetic tile target is completed by adding angle parameters.On this basis,different backbone networks are used for replacement and data sets with different proportions are divided for training.Through experimental comparison,the optimal network is determined to enable it to achieve accurate and rapid detection of magnetic tile targets at any angle. |
