9th International Conference on ICT for Sustainable Development

Authors - Ranispoorti Ravindra Patil, Jayalaxmi G N, V. H. Choudapur, Vishwanth P. Baligar
Abstract - Scanning Electron Microscopy (SEM) captures nanoscale images, whereas Deep Learning (DL) analyzes data with neural network models. The authors of this study investigate three types of SEM pictures at varying magnifications, and 93 pictures in all are created for processing and analyzing. SEM images are classed according to the doping component present in the main sample: Pure Cobalt Chromium Oxide, Lanthanum doped Cobalt Chromium Oxide, and Neodymium doped Cobalt Chromium Oxide. Augmentation methods increase the quality and look of images. The similarity and complexity of photos in all three groups, as well as the presence of images of varying magnification, offered a challenge to categorization, reducing the accuracy rate of the procedure. The authors suggest a low-complexity technique for extracting features and increasing model performance. In comparison with pre-trained models, this method reduces time and computing resources, increasing efficiency. Training accuracy is 100% while testing accuracy obtained is 78.26%. Additionally, CNN and classical deep learning models VGG16 and Inception v3 are used to classify SEM pictures. SEM images are segmented utilizing a holistic approach that incorporates the watershed algorithm along with contours. A SEM segmented image is used to determine the area of surface of a Cobalt Chromium Oxide sample. It was calculated that the material’s surface area was 41,02,628 nm2. . . .