PREDICTION OF THE SOIL PROPERTIES USING ARTIFICIAL INTELLIGENCE
Loading...
Date
2023
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
ICAR-INDIAN AGRICULTURAL RESEARCH INSTITUTE NEW DELHI-110012.
Abstract
ABSTRACT
Agriculture is pivotal for India, serving as the backbone of its economy by employing over 50% of the workforce. It ensures food security for its vast population and also contributes significantly to export revenues. Over time, agriculture has undergone transformations that unprecedentedly improved resource efficiency, yield, and profitability. As the world anticipates the "Agricultural Digital Revolution", the need for advanced, accurate soil mapping escalates. This revolution will ensure that agriculture meets the demands of the world's growing population. Soil testing in precision farming is paramount because it provides detailed insights into soil composition and health. This enables tailored agricultural practices, ensuring optimal nutrient application and irrigation. Consequently, it maximizes crop yield, reduces resource wastage, and promotes sustainable farming by aligning practices with specific soil needs. Soil pH and EC (Electrical Conductivity) testing are fundamental in precision farming. They provide critical insights into soil health and nutrient availability, enabling farmers to tailor agronomic practices for optimal crop growth, enhance nutrient utilization, and ensure sustainable soil management, ultimately leading to increased yields and cost-efficiency. Traditional soil testing involves manual collection and laboratory analysis, which is time-consuming, expensive, and provides coarser resolutions. Digital soil testing, leveraging sensors, offers rapid results and spatial variability insights in timely, scalable analyses. Addressing these limitations requires integrating the latest digital image processing and artificial intelligence (AI) technologies, particularly machine learning (ML) and deep learning (DL). Thus, This study harnessed the power of the machine and deep learning techniques, aiming to predict soil properties from pre-processed soil images. In this study, four AI models were investigated: Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Tree (DT), and Convolutional neural network (CNN) to train and test for classifying different soil pH and electrical conductivity (EC) classes. The K-nearest neighbors (KNN) model exhibited better performance in soil pH and EC prediction than the SVM and DT, achieving 61.9% and 65% accuracy, respectively. The model's AUC was 0.57 and 0.50, indicating that the KNN model could differentiate between the pH and EC classes, respectively, with a fair efficiency level. The accuracy of the CNN model for predicting soil pH was 65%, reflecting the overall correctness of all classifications made by the model. This is a decent level of accuracy but certainly leaves space for future improvements. The model's accuracy in predicting soil EC was impressively higher than pH
ii
prediction at 68.3%. This signifies the model's slightly greater competence in classifying soil EC correctly. While all the models provided useful insights into soil properties prediction, the CNN model, with its commendable precision, recall, F1-score, and accuracy in predicting both soil pH and EC, underscored the growing importance of machine learning in agriculture and soil management. However, there is a discernible space for future improvements in optimizing these models for enhanced prediction accuracy.
Keywords: Soil Properties, Soil pH, Soil EC (Electrical Conductivity), Image Processing, Soil Images, Artificial Intelligence, Machine Learning, Deep Learning (DL)