UAV Mission Height Effects on Wheat Lodging Ratio Detection

Wheat is an important staple crop worldwide, and lodging is a negative factor contributing to yield reduction. Obtaining timely and accurate wheat lodging information is critical. Using unmanned aerial vehicle (UAV) images for wheat lodging detection is a relatively new approach, in which researchers usually apply the manual method for field plot dataset generation. Considering the manual method being inefficient, inaccurate, and subjective, the study developed an image processing-based new approach for automatic field plot dataset generation and tested. Since only a few studies explored the effects of different UAV mission heights on wheat lodging ratio detection, we conducted experiments at three mission heights (15, 45, and 91 m) and analyzed images using machine learning (support vector machine—SVM) and deep learning (Resnet50) algorithms for wheat lodging ratios (3 grades) detection. The results indicated that the collected images on 91 m (2.5 cm/pixel) mission height could generate a similar or even a little higher detection accuracy over the images collection at 45 m (1.2 cm/pixel) and 15 m (0.4 cm/pixel) height. Comparison of SVM and Resnet50 model results showed that SVM resulted in more satisfactory performance (67.6% accuracy) with slightly higher accuracy over Resnet50 (67.2% accuracy) at 2.5 cm/pixel resolution. This study recommends that UAV images collected at the height of about 91 m (2.5 cm/pixel resolution), coupled with color and textural features and SVM classifier, is a useful approach for wheat lodging ratio detection. This study demonstrates the application of automatic plot extraction with low-resolution images produced by higher UAV mission height for plot scale (1.5 m × 3.6 to 14.6 m) studies such as wheat lodging forms a critical piece of information that could be adopted by users related to UAV applications and image analyses.