Classification of soybean leaf wilting due to drought stress using UAV-based imagery

Drought stress is one of the major limiting factors in soybean growth and productivity. Canopy leaf wilting (i.e. fast- and slow-wilting) is considered as an important visible symptom of soybeans under drought conditions. In soybean breeding programs, genotypes with the slow-wilting trait have been identified as drought-tolerant cultivars. Traditional method measures canopy leaf wilting traits using visual observations, which is subjective and time-consuming. Recent developments of field high-throughput phenotyping technology using Unmanned Aerial Vehicle (UAV)-based imagery have shown great potential in quantifying crop traits and detecting crop responses to abiotic and biotic stresses. The goal of this study was to investigate the potential use of UAV-based imagery in classifying soybean genotypes with fast- and slow-wilting traits. A UAV imaging system consisting of an RGB (Red-Green-Blue) camera, an infrared thermal camera, and a multispectral camera was used to collect imagery data of 116 soybean genotypes planted in a rain-fed breeding field at the reproductive stage. Visual-based canopy wilting scores were collected by breeders in the same day of imagery data collection. Seven image features were extracted, namely normalized difference vegetation index (NDVI), green-based NDVI (gNDVI), temperature, color hue, color saturation, canopy size and plant height for quantifying canopy wilting trait. Results show that all image features significantly (p-value < 0.01) correlated with soybean yield under drought. A Support Vector Machine model was developed to classify the two wilting traits using the images features and achieved an average classification accuracy of 0.8 with the highest one of 0.9. Slow-wilting genotypes had significantly (p-value < 0.01) higher NDVI, hue, saturation, canopy size, height, and lower temperature than fast-wilting genotypes. The significant broad-sense heritability (H-2) indicates the dominating genetic factors in the variations of the image features. The study demonstrates the good potential use of UAV-based imagery technologies in the selection of soybeans genotypes with drought tolerance.