Research On Machine Vision System Of Cigarette Box Defect Detection Based On YOLOv3 Algorithm

With the rapid development of intelligent manufacturing industry,the surface integrity of cigarette case becomes more and more important in the production process.At present,the traditional cigarette box surface defect detection is prone to miss detection and wrong detection.Based on the shortcomings of traditional methods,a cigarette case surface defect detection system based on deep learning and machine vision is studied.The main research contents are as follows.(1)This paper introduces the theoretical knowledge of deep learning and machine vision system,including image acquisition,image processing,image output and display.The neuron of neural network,the type of neural network model and the classical convolution neural network are introduced.Then,based on the theoretical basis,the network structure and loss function of YOLOv3 are improved.The network structure is increased by four times of down sampling,the Io U of loss function is changed into GIo U,and the candidate box is obtained by re clustering.The detection accuracy is improved by 7% compared with the original network,and the speed is slightly reduced.But it does not affect the detection in the industrial field.(2)In the hardware part,the light source,camera,lens and motor are selected.In the three-dimensional structure part of the platform,the dimensions of the light source mounting plate,camera mounting plate,lens fixing plate and motor fixing plate are described.In the software part,the camera image acquisition,camera positioning and motor positioning are described The call of the algorithm and the implementation of the interactive interface are described.(3)This paper introduces the data set and the environment of the experiment,then brings the data set into the convolution neural network for training,uses the model to detect,and shows the qualified rate of detection through the visual interface.