Machine Vision-Based Steel Defects And Logistic Position Recognition Research

In steel and iron enterprises,it's a critical issue to improve the quality and production efficiency of products.Due to various reasons,various defects will be inevitably produced on the surface of steel plates during the rolling process of steel plates.Timely identification of the defects plays an important role in the scheduling of production and the improvement of products quality.Steel plates with different specifications are usually stacked in the warehouse before the handling operation,and the accurate identification of steel plates can effectively improve the operation efficiency.Therefore,to study steel plate surface defect technology and the logistics location identification technology has great significance.The traditional steel surface defect recognition methods have some drawbacks,such as low recognition accuracy,poor real-time performance and possible introduction of new defects.The sensor is commonly used to identify the logistic location of steel plates,which is usually not accurate and can't identify the size of steel plates.Machine vision has a good performance on image recognition in the practical application.Therefore,this thesis studies the steel plate surface quality detection and the logistics location identification based on machine vision technology.1)In steel rolling process of steel production,the defects will decrease the quality of the steel and the problem should be addressed.This thesis introduces a method of the convolutional neural network based on differential evolution for the steel plate defect recognition.The optimization strategy of activation function selection of convolution neural network structure is proposed.Experiments show that reasonable configuration of the activation function can accelerate the training process of the convolution neural network model.A two-phase parameter optimization strategy based on differential evolution for convolutional neural network is proposed,which combines the local search ability of back propagation algorithm with the global searching ability of differential evolution algorithm for the parameters optimization.The experimental results show that this strategy effectively improves the recognition correct rate of convolution neural network.2)Aiming at the intelligent recognition of the steel plate logistics location information in warehouse operation,a location recognition method based on binocular vision for steel plate logistics is adopted in this thesis.The thesis proposes a hybrid image segmentation algorithm based on differential evolution algorithm and k-means algorithm for stereo matching algorithm,which improves the matching precision.The algorithm simulation experiment and the real physical system experiment show that this method is accurate and effective.3)With the background of actual demand,a steel plate defect recognition system based on improved convolution neural network is designed and developed.The system provides several functions of training model,image collection model,defect recognition model,image management model and so on.The system also provides users with the main functions of model training,image acquisition,defect recognition,image management and so on.Through the application of practical problems,it is shown that the system has the characteristics of simple operation and strong practicality.