Soil Moisture Detection Using Arduino Sensor and ANN Prediction

Abstract

Smart irrigation systems are essential to detect the existing moisture content of soil, which regulates and controls the water supply to irrigation. The present study focuses on the on-board installation of soil moisture sensor with Arduino UNO platform to measure the moisture content of soil samples, which will facilitate in releasing of irrigation water. The present experimental study uses five uniform (poorly graded) soil samples of size d50 = 850, 600, 425, 300, and 150 µm and a non-uniform (well-graded) soil sample of d50 = 325 µm. A fourth order polynomial is fitted between the sensor reading and degree of saturation, which is related to second order polynomial between the degree of saturation and moisture content of the soil. The sensor readings are used to estimate the existing moisture content of the soil sample through the degree of saturation of the soil through these polynomials. A satisfactory similarity is found between degree of saturation and versus normalized sensor readings for all the cases of uniform and no uniform soil. Further, power equation is developed between the sensor reading and the moisture content of the soil with an R² value of 0.96. In addition, three machine learning prediction models ANN, KNN, and SVM were employed and compared. It is found that artificial neural network predicted the moisture content better than other predictors having prediction accuracy with R² = 0.981 for training and 0.985 for validation indicating as a good predictor as compared to KNN and SVM.