Integration of Convolutional Neural Network and Thermal Images into Soil Moisture Estimation

To plan the proper irrigation strategy for the maximum crop yield in precision agriculture, many research works have been done extensively on how to estimate the soil moisture content. As the population and availability of thermal infrared sensors grow, their abilities to recognize the heat signatures and small differences in soil temperature help to estimate the soil moisture precisely. Since Convolutional Neural Network (CNN) has made a great success on the image recognition, we propose a scheme that integrates the thermal images captured by the sensors mounted on drones along with in situ measurements of the farm area and a CNN-based regression model to estimate the soil moisture. This model is used to estimate the soil moisture content through the soil temperature represented by thermal infrared images. Our experimental results show that our model performs better than the typical Deep Neural Network (DNN) with the test datasets, which in turn provides the more general estimation.