Classification and identification of tempered flexural organization of spring steel based on improved GLCM-Resnet18

Metallographic structure is generally judged by professionals based on existing knowledge and work experience, and the judgment results are somewhat subjective.In recent years, convolutional neural network (CNN) in deep learning methods can learn complex features in original images, widely used in the field of image classification and recognition.However, CNN require a large number of sample training to achieve good prediction results.In order to make up for the shortcomings of the subjectivity of manual judgment, and the problem that the data sets for specific problems in the field of materials engineering are often small, this study uses the grey level co-occurrence matrix (GLCM) to count the texture features of the original image, and then uses the standard Resnet18, Resnet50 and improved Resnet18 frameworks for migration training to classify and identify the grey level co-occurrence matrix of the troostite structure, in order to solve the problem of small metallographic image data sets and realize deep learning modeling of small samples.Using 490 microstructure images of spring steel tempered troostite collected by professional technicians, and each level have 98 images.The grey level co-occurrence matrix is used to count its texture information, thereby obtaining the training data set.The experimental results on this dataset show that the classification accuracy of the improved GLCM-Resnet18 can reach up to 96.52%, the highest accuracy of GLCM-Resnet18 is 95.65%, and the highest accuracy of GLCM-Resnet50 is 90.72%.It can be considered that the improved GLCM-Resnet18 method has more precise training accuracy and can basically meet the requirements of industrial applications.