Steel Plate Defect Detection And Classification Identification Method

Steel production of China is already the world’s number one.In 2018,steel production has accounted for 50% of the world’s total production,but the proportion of low-end and crude steel is still large.In recent years,due to overcapacity,most of the steel industry is in a state of loss.Nowadays,in the dark night of steel companies,in the face of industry sluggishness,increased competition,and decreasing demand,the era of winning by quantity will never return.The only way to become a steel company to spend the night.In the process of steel plate production,defects are inevitable,and the detection of steel plate defects is an important way to control the quality of steel plates.Therefore,it is of great significance to study the detection and identification of steel plate defects.Based on machine vision and image processing,this paper builds a steel plate defect detection system to study the identification and classification of steel plate defects.The steel plate defect detection system includes image acquisition,image processing,feature extraction and classification and identification steps.The CCD camera is used to collect the defect image,and the actual demand and special background for the detection and recognition of the steel plate defect are improved.In order to improve the picture quality,the related algorithms involved in the filtering denoising process are improved,and the binary image of the surface defect of the steel plate is obtained through image preprocessing.The geometric,grayscale and texture feature parameters in the binary image are extracted to determine whether there is a defect.The BP neural network algorithm is used to classify the image to determine the type,size and contour of the defect surface.In the image preprocessing part,according to the characteristics of steel plate noise,an improved median filtering algorithm is proposed by analyzing and summarizing the filtering algorithm.Reduces the number of window calculations and increases the speed of calculation.The target noise can be accurately identified,so that only the noise is processed,and the edge information and details of the image are well preserved.In the image segmentation process,the Ostu algorithm is selected.The Ostu method is a method for automatically determining the threshold value by maximizing the variance between classes.The three values of foreground and background weight,gray mean and variance are calculated.And by calculating the sum of the specific gravity and the variance,the calculated value of the variance between the images under the threshold is obtained,and the target and the background are well distinguished.In the feature extraction process,feature information is selected for geometric features,gray features and texture features,and then the obtained data information is standardized and PCA dimensionality reduction processing.In the classification and recognition part of steel plate defects,four kinds of defect samples were selected,the total number of samples was 1200,800 were selected as training samples,and the remaining samples were used to test the classification results.The sample eigenvalues are used as the input values of the BP neural network.After the network learning,the test results are obtained.According to the specific defect type comparison of the samples,the validity of the BP network classifier is verified.By comparison,the classification works well.