The Research And Application Of Detection Algorithm For Surface Defects On Mobile Phone Lithium Battery

With the rapid development of smartphones,the production demand of mobile phone lithium battery is on the increase day by day.In order to ensure the production efficiency of mobile phone lithium battery,the manufacturing process of lithium battery has implemented basically unmanned operation,but its appearance quality inspection is still in the stage of manual inspection.Therefore,the introduction of automatic detection method based on machine vision has become a research hotspot in industrial detection.In the process of machine vision detection,due to the complex production process,lighting conditions and other factors,the quality of the collected lithium battery appearance image is different,resulting in low detection accuracy with traditional machine vision methods.Therefore,in this thesis,we studied the algorithm and application of defects detection on mobile phone lithium battery surface,proposing a significance enhancement algorithm based on image difference to extract the feature information of lithium battery surface defects,and building an object detection model based on the deep learning to realize the classification and location of lithium battery surface defects.In addition,we also developed a set of industrial visual inspection software based on deep learning and applied the built detection model to the software system.For feature extraction problem of defects with complex illumination,we proposed a method based on image difference in this thesis,which uses two images with different illumination schemes as difference operation to suppress the noise problem of uneven illumination,and enhances the defects feature information based on significance algorithm and the first-order discrete Gauss differential.Experimental results show that the algorithm can effectively enhance the feature information of surface defects and get the defect location after edge extraction.However,this method has two problems:the defect with weak feature information is difficult to detect,and the overall detection speed is slow.In order to further improve the detection effect of surface defects of lithium battery,a ConcateXnet model based on grouped convolution with concatenate blocks was proposed in this thesis.The model is simple in structure and small in calculation,which can improve the speed of training and testing.At the same time,it can effectively extract the features of defects and avoid the loss of feature information by using the way of cascading and concatenating with the shallow and deep feature maps.According to the actual hardware and software requirements of the industrial site,we designed a three-light-source camera module and developed a set of industrial visual inspection software based on deep learning in this thesis.The hardware module can collect images from the left,right and vertical side under three different directions of the three-light-source camera module,which can enhance the imaging effect of defects.The software system integrates function modules such as model training and testing,remote deployment and model calling.In this thesis,the constructed lightweight network model can be applied to the software system,and 294 pictures of lithium battery surface collected in the field are manually marked with defect location and category for model training and testing.The actual testing results show that the ConcateXnet model can realize end-to-end real-time detection,and obtain detection speed with 26 fps and average accuracy(map@0.5 IOU)with 92%,which meets the actual detection speed and accuracy requirements of the experimental environment and industrial site.