Intelligent Detection Of Tire Defects Under X-ray Based On Deep Learningg

X-ray defect detection technology for automobile tires is of great significance in terms of traffic safety and tire production.With the continuous update of detection technology,tire X-ray defect detection technology based on deep learning has also been developed by leaps and bounds.Although many inspection experts at home and abroad have put forward many ideas for tire defect detection,the internal structure of tires is too complicated and there are too many types of tire defects,so there is no method that can be used in industrial production inspections.The detection technology based on deep learning has a long research history,but most scholars use supervised learning methods.This method requires too much sample labeling,too many influencing factors,and a single algorithm cannot be applied.The defect types of all tires,In order to solve the problems of large amount of supervised learning,inability to analyze the spatial distribution of samples,and inability to identify multiple defects,a semi-supervised learning cluster generation confrontation network structure is constructed,which is better suitable for the detection of multiple defects in tire X-rays.In this paper,further research is conducted on tire X-ray defect detection based on deep learning:(1)Two aspects are introduced to the research status of tire inspection at home and abroad,including methods based on deep learning and traditional pattern recognition,and a detailed introduction to the structure and defect types of tires.The samples in this article come from the databases of major domestic manufacturers.The training set is produced uniformly,so data preprocessing methods such as sample labeling,sample cropping,sample noise reduction,and sample expansion are required.(2)Construct a semi-supervised learning network—the clustering generation confrontation network.The network is composed of a pooling layer,a convolutional layer,a fully connected layer,and an activation function layer.The network structure is divided into three parts: generator,discriminator and decoder,which is a better feature extraction method for tire X-ray images.(3)The current optimization algorithms are compared and demonstrated,and the momentum-based gradient descent optimization algorithm is determined to adjust the network parameters and update weights,thereby improving the recognition accuracy of the network and the effect of generating pictures.(4)The clustering research of cluster generation confrontation network,based on the generation confrontation network,accurately reconstruct the latent feature vector through the encoder,and determine the Gaussian mixture model clustering algorithm through the multi-faceted study of the clustering method,which can Spatial clustering.The deep learning platform used in this article is Pytorch.Both the training set and the test set are composed of homemade tire X-ray samples.The recognition accuracy of tire background samples reaches 100%.The spatial distribution of unrecognizable samples is analyzed through the cluster graph after dimensionality reduction.It also improves the diversity of the network while ensuring the recognition rate.