Research And Implementation Of R Angle Of The Tab On Lithium Battery Localization Algorithm In Complex Background

In recent years,the new energy industry continues to develop.As the core component of new energy products,lithium battery is widely used in automobile,mobile phone,medical products,military equipment,aerospace and other industries.The R angle of the tab plays an important role in the production process of lithium battery,which uses the R angle of the tab position in the cutting machine to locate the cutting knife,in the die cutting machine and slitting machine through the R angle of the tab to find the tab area and measure the tab width and height.With the continuous increase of production speed,the enterprise’s requirements for the positioning accuracy of the R angle of the tab on lithium battery are also improved accordingly.The traditional image processing method have some problems for the R angle of the tab on lithium battery in complex background,such as low positioning precision and poor universality,only by accurately locating the position of the R angle of the tab can ensure the accuracy of the cutting knife cutting and measuring accurately,reduce the waste of raw materials,Therefore,higher requirements are required for positioning algorithm of the R angle of the tab.In view of the above problems,the thesis adopts deep learning method to locate the R angle of the tab on lithium battery under complex background.Firstly,datasets required for model training and testing were constructed.Secondly,four classical networks were selected for comparative experiments,and the initial network structure was determined as Faster RCNN.Based on the shortcomings of the model,an improved Faster RCNN network was proposed.Finally,a real-time positioning program for the R angle of the tab on lithium battery is developed.The main research contents are as follows:（1）Collate and analyze the collected data set.Due to the lack of complex background images of samples,insufficient number of samples and large pixels of the original image,data enhancement of samples is required,and methods such as increasing noise,image flipping,clipping and compression are adopted to enhance the data set.（2）Four classical target detection networks were used to conduct experiments on the R angle of the tab on lithium battery.By comparing the experimental results,Faster RCNN was selected as the initial network structure.The network is optimized for problems such as weak feature extraction ability of Faster RCNN network,unrepresentative setting of Anchor,and certain location error of ROI Pooling layer.（3）Improve the feature extraction network of the Faster RCNN network.Four convolutional neural networks,VGG16,Mobile Net V2,Res Net50 and Res Net34,were compared in experiments.Res Net34 was selected as the feature extraction network of Faster RCNN according to the two performance indexes of average precision and detection time.In order to solve the problem of poor localization accuracy of small pixel R angle of the tab,the feature fusion network is introduced to realize multi-scale information fusion and improve the localization ability of the network for small target.Experimental results show that this method improves the localization accuracy of small target.（4）Improve RPN network of Faster RCNN network.The detection ability of the original RPN network is weak,so the convolution layer is added to the RPN network to increase the receptive field and improve the detection ability of the network.At the same time,k-means algorithm was used to cluster the scale and aspect ratio of Anchor of the R angle of the tab on lithium battery,and 15 anchors suitable for the dataset of the thesis were obtained and applied to RPN network.Experimental results show that this method can further improve the positioning accuracy of the network.（5）Adopt ROI Align in the ROI Pooling layer to reduce positioning error.ROI Align cancels two integer operations,directly reserves the position coordinates to decimal places,and finally obtains the eigenvalues of floating point coordinates through bilinear interpolation.Experimental results show that this method improves the positioning accuracy of the network.（6）The development of a program for real-time locating the R angle of the tab on lithium battery.Using Python and Qt mixed programming method for the R angle of the tab on lithium battery positioning program interface development,can realize the detection of a single image and real-time display of positioning frame,to intuitively display positioning function for the R angle of the tab on lithium battery.The experimental results show that the improved network model can detect R angle of the tab on lithium battery with an average precision of 97.2%,the average positioning error is 3 pixels,and the detection time of a single image is 36 ms,which meets the requirements of R angle of the tab on lithium battery in the factory.