Unsupervised discrimination between lodged and non-lodged winter wheat: a case study using a low-cost unmanned aerial vehicle

摘要

Wheat lodging significantly reduce the grain yield and quality. Mapping wheat lodging timely and accurately can help farmers get full compensatory damages in time. Therefore, the objective of this study was to employ a low-cost unmanned aerial vehicle (UAV) carrying a Red-Green-Blue camera to discriminate lodged from non-lodged wheat effectively. Low-cost UAV is easier to be accepted by Chinese farmers most of whom manage small-scale farms. After comparing a variety of colour features as well as their texture features, this study found that the texture feature of mean of G/B (digital number ratio of the green band to the blue band) derived from occurrence measures was the optimum discriminator of lodged and non-lodged wheat. This discriminator was still effective although the spatial spectral variations in the study area were much more complex than that in previous studies. With an unsupervised method based on the discriminator, the UAV system was able to discriminate lodged wheat from non-lodged wheat. The resultant of overall accuracy was 89.55% and Kappa coefficient was 0.76. The producer's accuracies were 81.23% and 93.62%, whereas the user's accuracies were 86.15% and 91.08% for lodged and non-lodged wheat, respectively. The retrieved wheat lodging distribution map showed a substantial agreement with ground reference data.