Soil Class Ensemble Approach Using Deep Learning Models

Agriculture serves as the main source of livelihood of a significant portion of the Indian population and thus is a major economic sector of the country. One of the most significant aspects of agriculture is the classification of soil types, which essentially helps in gaining the knowledge about the composition and characteristics of soil. Soil classification through conventional laboratory methods is a tedious, expensive, and highly specialized procedure. We considered five different soil types such as Alluvial, Black, Sand, Gravel, and Loam have been predicted using a very low cost technique by analyzing a total number of 1269 soil images comprising data.Various pretrained deep neural network models such as DenseNet121, VGG16, Xception, MobileNet, and InceptionV3 were used to carry out the investigation in depth. Accuracies of 90. 15