Integrating thermal infrared and RGB imaging for early detection of water stress in lettuces with comparative analysis of IoT sensors

Abstract

Early water stress detection is important for water use yield and sustainability. Traditional methods using the Internet of Things (IoT), such as soil moisture sensors, usually do not provide timely alerts, causing inefficient water use and, in some cases, crop damage. This research presents an innovative early water stress detection method in lettuce plants using Thermal Infrared (TIR) and RGB images in a controlled lab setting. The proposed method integrates advanced image processing techniques, including background elimination via Hue-Saturation-Value (HSV) thresholds, wavelet denoising for thermal image enhancement, RGB-TIR fusion using Principal Component Analysis (PCA), and Gaussian Mixture Model (GMM) clustering to segment stress regions. The leaves stressed areas annotated in the RGB image through yellow pseudo-coloring. This approach is predicated on the fact that when stomata close, transpiration decreases, which causes an increase in the temperature of the affected area. Experimental results reveal that this new approach can detect water stress up to 84 h earlier than conventional soil humidity sensors. Also, a comparative analysis was conducted where key components of the proposed hybrid framework were omitted. The results show inconsistent and inaccurate stress detection when excluding wavelet denoising and PCA fusion. A comparative analysis of image processing performed on a single-board computer (SBC) and through cloud computing over 5 G showed that SBC was 8.27% faster than cloud computing over a 5 G connection. The proposed method offers a more timely and accurate identification of water stress and promises significant benefits in improving crop yield and reducing water usage in indoor farming.