Semantic Image Segmentation using CNN (Convolutional Neural Network) based Technique

Segmenting images makes computer vision applications more manageable. By breaking an image into smaller groups called “image segments,” an image segmentation approach simplifies the complexity of an image for later processing or analysis. Individual pixels are labeled as part of the process known as segmentation. Many computational research, modeling, and forecasts require a sufficient microstructure sample. An alternate method is to extrapolate a 2D image into a virtual 3D sample, which saves time and money compared to collecting 3D microstructural images. A CNN Architecture-based U-NET technique was used to find extrapolation-based reconstructions for alloys, porous media, and sandstones. Using a deep convolutional neural network, VGG16, the objective function in this optimization is built using microstructural properties. We use a permutation operator to create multiclass feature maps from these stacks. Sandstone datasets, which contain a number of unique materials, are used to demonstrate the actual application of our technique. Pre-annotated data was used to guide our analysis. Image reconstruction is a commonly used approach in various fields for deciphering the many objects in a picture. Using U-NET, we first segment the photographs and then reassemble the image slices to be shown. The overall IoU and accuracy rates for this suggested design were both high, at 85% and 96%, respectively.