Deep learning based multi-labelled soil classification and empirical estimation toward sustainable agriculture

Agriculture is the underlying occupation of the vast people in India and it is a major economic contribution. Soil is prime for the vital nutrient supply to the crops and its yield. Determination of the type of soil which comprises of the clay, sand and silt particles in the respective proportion is indeed significant for the suitable crop selection and to identify the weeds growth. The most commonly utilized soil determination methods were International Pipette method and Pressure-plate apparatus method. In this research work, multiclass soil classification using machine learning and deep learning models for the appropriate determination of the soil type as Multi-Stacking ensemble model and a novel feature selection algorithm Q-HOG is proposed; since the Artificial Intelligence has led to furtherance in the smart agriculture. Besides, the images are collected from the exploration site vriddhachalam along with the soil datasets will increase the classification accuracy. The deep learning models Recurrent Neural Network(RNN), Long Short Term Memory(LSTM), Gated Recurrent Unit(GRU) and VGG16 are considered and the comprehensive evaluation of these different deep learning architectures and also the machine learning algorithms such as Naïve-bayes, KNN, SVM are carried out and the obtained results are tabulated. Multi-stacking ensemble model for multi-classification is proposed with the Machine learning and deep learning algorithms and evaluated the performance with increased computation time. Among these models the proposed model outperformed in soil classification in-terms of accuracy as 98.96 percent, achieved precision as 96.14 percent, recall as 99.65 percent and the achieved F1-Score is 97.87 percent.