Prototype and Performance Evaluation of a CNN Model for Soil Moisture Estimation in Wet Andosols.

Abstract

Soil moisture is an important factor that determines the workability of agricultural machinery and crop growth. However, existing remote sensing methods by which to estimate soil moisture are adversely affected by both the external environment at the time of imaging and the high costs of sensors. Hence, in this study, we develop a Convolutional Neural Network （CNN） model for soil moisture estimation. The proposed model uses visible light images to estimate soil moisture under wet conditions with high accuracy. Moreover, we evaluate the performance of the proposed model in the establishment of an inexpensive remote sensing method for soil moisture that is less susceptible to external environmental factors. Therefore, the proposed model is an inexpensive, simple, and environmentally independent method by which to estimate soil moisture, compared with existing remote sensing methods.