An Efficient Optimization Approach for Steel Surface Flaw Classification using Machine Learning

Automatic steel surface defect classification is a well-known problem and is being considered for more than two decades. On the other hand, manual methods for defect classification in steel products are time consuming, labor intensive, and are prone to error. As a result, a real-time surface defect classification method using genetic algorithm is proposed to classify four kinds of surface defects in steel surfaces such as Roller Mark, Rust Spot, Emulsion Spot and Scrape. The proposed inspection system consists of visual dataset of 500 steel defect images for processing. The overview of the classification model has online and offline process. In offline, the input image is pre-processed and the following visual features such as geometric features, shape features and texture features are extracted from the defect image. In order to optimize the extracted visual features and to improve the accuracy of the classification model, genetic algorithm is used to find an optimal solution. For optimization, an initial population is generated randomly and the fitness function is calculated for every chromosome. Then the genetic operations such as selection, crossover and mutation are done to generate new offspring chromosome and evaluated to pick optimal chromosome or optimal feature. Finally, in online, the selected features are given as input to the Optimized SVM classifier, in order to predict the class of defect to which the input image belongs. Accuracy, sensitivity, specificity, f-measure and precision are used as the performance metrics to evaluate the system.