Classification of Different types of Soil using Deep Learning

The present study aims to add a new dimension in the classification of soil using deep learning in conjunction with the conventional methods. In this regard, a CNN model has been developed and trained using a small, publicly available dataset for classifying five specific classes of soils. The proposed model had an accuracy as high as $\mathbf{9 0. 5 2 \%}$, proving the efficiency and swiftness of the model for soil classification. In future works, the dataset will be improved with crowdsourcing to increase diversity and better coverage. The un-annotated status for the dataset can be relaxed with online data augmentation techniques. This study will help develop a completely automatic, efficient method for soil classification, which has broad applications in agriculture, geotechnical engineering, and environmental monitoring.