SOIL TYPE CLASSIFICATION FROM HIGH RESOLUTION SATELLITE IMAGES WITH DEEP CNN

The primary focus of the method, developed here is on classifying soil types from satellite images with deep convolutional neural network. As we know there exist different types of soil around the land regions of the earth. But the physical and chemical properties of soil varies not only with change in regions, but also with time due to the factors like, deforestation, addition chemical waste etc. Here in this work we aim to develop a method to classify different types of soil based on their physical and chemical properties by analysing satellite images. Here four primary soil types, namely, alluvial, black, desert and red, available in India has been considered for our experimentation. An image repository, with around 1000 images, labelled with their soil-types, is also created for this work. The images, used here are of high resolution, acquired with LANDSAT-8 satellite. We have performed our study with these large number of labelled images from each soil type, captured in different time. We developed a method of soil classification with deep convolutional neural network. The merits of this method is experimentally validated with suitable comparison and proved to be effective in identifying proper soil class accurately.