Segmentation of Terrain Classes from Satellite Images using Modified U-Net Algorithm

A substantial number of computerized agricultural models are aimed at assisting ranchers in monitoring their farms throughout the season. For horticultural applications, knowing the exact harvest limitations is very important. Nonetheless, this task is mostly manual and consumes more time. Mechanizing the field extraction works will not only limit the movement of ranchers, encouraging a more extensive selection of these administrations, but it also allows for supplying better product and administrations to be supplied by employing far off detection. Convolutional Neural Networks (CNN) became additional vogue in image analysis due to their ability to automatically learn the relevant discourse choices. At the start it is devised for natural images, these networks' square measure revisited and made-to-order to tackle remote-sensing problems like road extraction, cloud detection, crop identification, water body, and concrete mapping. Thus, they appear to be quite viable to extricate field limits. Our general point is to create and pass judgment on the easiest method to remove field limits at scale by each substitution setting explicit discretional picks and lessening the picture pre technique work. This methodology extricated field limits with high topical and mathematical precision. This paper attempts to envision the presentation and impediments of the projected methodology utilizing deep learning. The proposed model was able to predict various classes with an accuracy of 92% for training of 8 EPOCHS. The validation accuracy was 82%. The model could successfully differentiate between terrain classes like roads, buildings and fields and could even find classes that were not present in the ground truth data.