Detection of Biotic or Abiotic Stress in Vineyards Using Thermal and RGB Images Captured via IoT Sensors

Abstract

The abiotic and biotic plant stress is a critical factor for the agriculture industry, considering that plant stress increases production costs and reduces the quantity and quality of the output product. The early detection of plant stress using new technologies and even more utilizing Internet of Things (IoT) sensors is significant for the growers to act as early as a problem is in the nascent. Furthermore, the insights from the early detection of plant stress can be used as actionable data for fertilization and pesticide optimization. Our proposed state-of-the-art method uses Thermal Infrared (TIR) and high-resolution visible-spectrum (RGB) images acquired by IoT sensors of Unmanned Aerial Vehicles (UAVs) from two experiment vineyards (Vitis vinifera L.) for two years in a total of twelve flights. The OTSU method is used for the plants' canopy isolation from the soil. The k-Means clustering is used in the relative temperature values of the plant's canopy to detect the leaves' stomatal closure. The clusters' pixel coordinates of the TIR image, which represents leaves' stressed areas, are used, and a pseudo-coloring of yellow is assigned in the corresponding pixels of the aligned RGB image. Finally, an RGB image is generated with yellow pseudo-coloring over the stressed areas of the vineyards' canopy. The stressed plants caused by abiotic and biotic factors were validated and compared with the Triangular Greenness Index (TGI), which measures the leaf chlorophyll content like other multispectral indexes. Finally, the proposed method shows significantly higher accuracy and precision than TGI based on the two years of experimentation results, considering that the F1-score of the proposed method is better than TGI in the cumulative of the two years by 70.55%.