Fig. 1

Workflow of wheat plot lodging ratio detection under different mission heights for unmanned aerial systemsNote： Non-， light， and severe lodging represent 0%， <50%， and >50% crop lodged， respectively； SVM—support vector machine； RF—random forest； KNN—K nearest neighbors； three heights are 15， 46， and 91 m

Fig. 2

The location of field trials （left） and field trials layout （right）Note： On the right image， columns 1 （left most）， 11， 20， 29， and 55 （right most） are border plots

Fig. 3

General procedure on the development of automatic plot image generationNote: (a) original image； (b) monochrome of excess green； (c) binary image after thresholding (0.01) excess green image (b)； (d) relationship between x coordinates (horizontal direction) and sum of each column； (e) relationship between y coordinates (vertical direction) and the sum of each row； and (f) samples of automatically generated extracted images of the individual plots

Table 1

Features selected based on domain knowledge for wheat lodging ratio detection

Fig. 4

Detection accuracy of different wheat lodging ratios based on color and textural features extracted from unmanned aerial systems images collected at three flight altitudes coupled with SVM， random forest （RF and KNN classifiers）Note： Bars for each group with different letters are significantly different by Tukey test （α = 0.05）

Fig. 5

Detection accuracy of different wheat lodging ratio based on color and textural features extracted from unmanned aerial systems images collected at three flight altitudes coupled with RF classifierNote： Bars for each group with the same letters are not significantly different by Tukey test （α = 0.05）

Fig. 6

Detection accuracy of wheat lodging ratio based on color and textural features extracted from unmanned aerial systems images collected at three flight altitudes coupled with ResNet100， GoogLeNet and VGG16 deep learning models

Table 2

Detection accuracy comparison between RF and ResNet101 at different flight heights Unit： %