Research On The Recognition And Classification Of Glass Bottle Defects And Bottle Bottom Model Number

Glass bottle quality inspection is an important part of the bottle making industry.Quality control is related to customer needs,and data statistics help process improvement.The glass bottle forming process is under high temperature conditions(up to 1500 degrees Celsius),and the process control is very difficult(8%-10% failure rate),so quality inspection is particularly important.Through on-site inspection of the quality inspection on the production line,it is found that the defects on the glass bottle are small and difficult to distinguish,especially the defects located at the mouth of the bottle.The classification accuracy of the model number is not high,so an improved method is proposed: based on the convolutional neural network,the weak feature defects of the glass bottle and the model number of the bottom of the bottle are classified,and the visual interface design of images and data is realized.The main contents are summarized as follows:(1)Pre-process the image set of defects and bottle bottom mold numbers.This article is aimed at Qianhe soy sauce bottles.The research objects are 5 types of bottle mouth images(4 types of defects and 1 type of normal bottles)and 5 types of bottle bottom model numbers.The images are from the photo library of the current camera inspection machine.By rotating,translating,and flipping the collected image sets,the number of expansions is increased(10,000 images per category);contrast enhancement and bilateral filter denoising methods are used to eliminate interference in the defect image set;mean and normalization are performed on the defect image set In one operation,the pixel value is centered and the value is mapped to 0 ? 1 to improve the convergence speed of the training network.(2)Build and optimize the convolutional network of bottle bottom model number and bottle mouth defect.Based on the convolutional network model,gradually build and improve the bottleneck model number classification convolutional network,and finally reach 98.95% classification accuracy,and compare with the three types of classic networks to evaluate the performance of the network.Transfer the model number classification network structure to defect training,and gradually adjust the activation function and parameter update method in the original structure according to the changes in the accuracy curves of the training set and verification set in the experiment,adjust the dimensions of the convolution kernel,Batch?Size value,number of iterations,Learning rate and other parameters,and introduce the Dropout layer and BN(Batch Normalization)layer,and finally add Inception and Res Net structures to the network structure,and train the weight of the bottle bottom model number image set,and the five types of image sets in Fashion-MNIST training the final weight is used as the initial value of the weight of the defect classification network.After the defect network training is completed,the final accuracy reaches 97.86%.Comparing the network with two types of classic networks,the results show that the classic network also performs better in defect classification,but the network built is more concise and realizes the classification task with the least number of layers and better performance.(3)The deployment of glass bottle bottom mold number and bottle mouth defect classification system is mainly divided into image acquisition module,image classification module,data storage module,and visual display module: obtain image data through the camera,and input it into the trained module number and defect classification convolution Network model.Store the classification results in the built database.Use the Tkinter library provided by Python to complete the front-end design of the interface and realize the real-time display of detected images and data.Connect and build a database for storing test results based on Python,and use the Matplotlib library provided to achieve statistical chart drawing.Based on the "We Chat Developer" software platform,the We Chat Mini Program is designed to facilitate data viewing on the mobile terminal.Finally,the system software was tested to achieve the classification effect.