Application of Deep Learning to Improve the Accuracy of Soil Nutrient Classification

Soil characteristics are required to calculate of computer soil nutrients for specific crops productivity. Soil testing is necessary for determining the soil’s nutritional level determining the treatment of the soil for its fertilities and crop requirements. Soil test reports are used to classify the significant soil features by indices of Phosphorus, Potassium, Organic carbon, and Boron for parameter soil reaction (pH). Improved accuracy in prediction of the soil parameter will aid to reduce expenditure on fertilizers will increase farmer’s profit. Machine learning comes under AI (Artificial Intelligence), and a superset of Deep Learning. Deep Learning mostly uses CNN (Convolutional Neural Network) which is an extended and inspired architecture to make classification and predictions with more percentage of accuracy. In this paper accuracy of classification for soil nutrients are analyzed and improved using deep learning methods such as CNN, which gives better accuracy results. Image classification can be done using regression and SVM (support Vector Machine) also but CNN from Deep Learning model is more effective in learning patterns in 3-dimensional data of colour images and by automatic feature learning capability in the proposed CNN model. Images of soils are used to train a classifier to identify common soil types and fractions of soil type. OpenCV and TensorFlow tools of Python are used for image processing, testing and training soil image data. Using these tools images of 3 soils (gravel, sand and silt) of Karnataka region of India have been taken for training and testing model and classification accuracy is improved.