Research On Classification Algorithm Of Magnetic Ring Surface Defects Based On Machine Learning

With the rapid development of the Internet and digital technology,we are increasingly researching areas such as artificial intelligence,machine learning,and automatic identification.Machine learning is the fundamental way to make computers intelligent,and its applications are all over the field of artificial intelligence.As a competitive machine learning algorithm,Extreme Learning Machine(ELM)attracts more and more scholars' attention with the characteristics of its simple theory,strong generalization ability and fast learning speed.In the classification and regression issues,ELM has been widely used and achieved good results.For the classification problem of magnetic ring defects,traditional artificial visual has been unable to meet people's needs for accuracy and real-time.The classification methods based on image processing increasingly being applied to all aspects of industrial production.In this context,this paper uses the ELM and other machine learning algorithms to study the defect classification of the magnetic ring surface.In the paper,the image of the magnetic ring is obtained by the area CCD camera plus microphotography,then we briefly analyzes the common magnetic ring defects in production,designs the general process of defect classification,and compares the classification algorithms according to the classification accuracy and time requirements.The details are as follows:For the image enhancement,this paper compares three classic gray transform methods firstly and selects the linear transformation with ideal transformation effect to enhance the image contrast.Then,the Gaussian filtering method is adopted after comparing several classic filtering methods,and the evaluation results show that the filtering noise is better.For the image segmentation,this paper briefly studies the edge detection and threshold segmentation methods,but they don't get expected effect on the segmentation of magnetic ring defects.In this case,the multi-scale Gaussian filtering combined with OTSU segmentation method is designed.It is found that the segmentation effect for deep and long scar defects is ideal,but the effect for cracks,air bell and other defects is general.Then,multidimension and multi-directional 2D Gabor filters segmentation method is proposed,because it can smooth the defect image from different dimensions and directions,and can determine the segmentation threshold automatically according to the gray level,which greatly improves the adaptability and accuracy of segmenting various defects.For the defect classification,this paper classifies the magnetic ring surface defects by machine learning algorithms,such as BP neural network,Support Vector Machine(SVM)and ELM.The feature vector after feature extraction and dimensionality reduction is used as the model input of the classification algorithm,then the model is trained and predicted.After determining the key parameters of the activation function and the number of neurons of the ELM classifier.It is confirmed that the ELM can meet the requirements of rapidity and accuracy.On this basis,the KELM operating rate and stability combined with the RBF kernel function are further improved.Compared with the former two methods,the magnetic ring defect classification algorithms based on ELM and KELM are proved to have good practicability.