Using continous wavelet analysis for monitoring wheat yellow rust in different infestation stages based on unmanned aerial vehicle hyperspectral images
Qiong Zheng, Wenjiang Huang, Huichun Ye, Yingying Dong, Yue Shi, and Shuisen Chen
1Key Lab of Guangdong for Utilization of Remote Sensing and Geographical Information System, Guangdong Open Laboratory of Geospatial Information Technology and Application, Research Center of Guangdong Province for Engineering Technology Application of Remote Sensing Big Data, Guangzhou Institute of Geography, Guangzhou 510070, China
2Key Laboratory of Digital Earth Science, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China
3Key Laboratory for Earth Observation of Hainan Province, Sanya 572029, China
4Department of Computing and Mathematics, Faculty of Science and Engineering, Manchester Metropolitan University, Manchester M1 5GD, UK
Yellow rust is the most extensive disease in wheat cultivation, seriously affecting crop quality and yield. This study proposes sensitive wavelet features (WFs) for wheat yellow rust monitoring based on unmanned aerial vehicle hyperspectral imagery of different infestation stages [26 days after inoculation (26 DAI) and 42 DAI]. Furthermore, we evaluated the monitoring ability of WFs and vegetation indices on wheat yellow rust through linear discriminant analysis and support vector machine (SVM) classification frameworks in different infestation stages, respectively. The results show that WFs-SVM have promising potential for wheat yellow rust monitoring in both the 26 DAI and 42 DAI stages.
You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
You do not have subscription access to this journal. Figure files are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
You do not have subscription access to this journal. Article tables are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
You do not have subscription access to this journal. Equations are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
${R}$ represents band reflectance. The spectral sampling interval of S185 was 4 nm, and the bands centered in 680 or 800 did not existed in its band settings. Therefore, the reflectance of the band that is not in the band setting was calculated as an average of the reflectivity of the adjacent two bands (e.g., the reflectance of bands 680 or 800 in the S185 data were calculated as the average reflectance at 682 and 678 nm and at 802 and 798 nm, respectively).
Table 3.
Optimization of Vegetation Index Based on S185 Bandsa
Indicates the significant correlation in ${p}$-value ${\lt}\;{0.05}$.
Indicates the significant correlation in ${p}$-value ${\lt}\;{0.01}$.
Indicates the significant correlation in ${p}$-value ${\lt}\;{0.001}$.
Table 4.
Location and Scale Parameters of Sensitive Wavelet Features in Different Infestation Stages
Stages
Features
Range of Spectrum
Location (nm)
Scale
26 DAI
WF1
Green
474–546 (530)
0.76
WF2
Yellow
562–594 (574)
0.71
WF3
Red
574–682 (674)
0.74
WF4
Red-edge
706–710 (706)
0.67
WF5
Near infrared
754–762 (758)
0.67
42 DAI
WF1
Green
490–526 (510)
0.79
WF2
Red
618–678 (662)
0.82
WF3
Red-edge
694–726 (714)
0.85
WF4
Near infrared
762–774 (766)
0.82
WF5
Near infrared
786–794 (790)
0.82
Table 5.
Classification Accuracies of LDA and SVM Models Based on the Vegetation Index Features in 26 DAI Stagea
Methods
Training Data
Validation Data
LDA
healthy
infection
U (%)
OA (%)
Kappa
healthy
infection
U (%)
OA (%)
Kappa
healthy
134
39
77.4
85.7
0.69
62
12
83.8
86.7
0.70
infection
26
256
90.8
18
133
88.1
(%)
83.8
96.8
77.5
91.7
SVM
healthy
infection
U (%)
OA (%)
Kappa
healthy
infection
U (%)
OA (%)
Kappa
healthy
132
23
85.2
88.8
0.75
65
12
84.4
88.0
0.74
infection
28
272
90.1
15
133
89.9
(%)
82.5
92.2
81.3
91.7
For Table 5–8: P = producer’s accuracy, U = user’s accuracy, OA = overall accuracy.
Table 6.
Classification Accuracies of LDA and SVM Models Based on the Vegetation Index Features in 42 DAI Stage
Methods
Training Data
Validation Data
LDA
healthy
infection
U (%)
OA (%)
Kappa
healthy
infection
U (%)
OA (%)
Kappa
healthy
141
11
92.8
93.4
0.85
72
9
88.9
92.4
0.84
infection
19
284
93.8
8
136
94.4
(%)
88.1
96.3
90.0
93.8
SVM
healthy
infection
(%)
OA (%)
Kappa
healthy
infection
(%)
OA (%)
Kappa
healthy
153
15
91.7
95.2
0.95
80
6
93.0
97.3
0.94
infection
7
280
97.6
0
139
100
(%)
95.6
94.9
100
95.8
Table 7.
Classification Accuracies of LDA and SVM Models Based on the Continuous Wavelet Features in 26 DAI Stage
Methods
Training Data
Validation Data
LDA
healthy
infection
(%)
OA (%)
Kappa
healthy
infection
(%)
OA (%)
Kappa
healthy
125
21
85.6
87.7
0.72
64
14
82.1
86.7
0.71
infection
35
274
88.7
16
431
89.1
(%)
78.1
92.9
80.0
90.3
SVM
healthy
infection
(%)
OA (%)
Kappa
healthy
infection
(%)
OA (%)
Kappa
healthy
129
14
90.2
90.1
0.78
68
9
88.3
90.7
0.79
infection
31
281
90.1
12
136
91.9
(%)
80.6
95.3
85.0
93.8
Table 8.
Classification Accuracies of LDA and SVM Models Based on the Continuous Wavelet Features in 42 DAI Stage
Methods
Training Data
Validation Data
LDA
healthy
infection
(%)
OA (%)
Kappa
healthy
infection
(%)
OA (%)
Kappa
healthy
153
15
91.1
95.2
0.90
75
7
91.5
94.7
0.88
infection
7
280
97.6
5
138
96.5
(%)
95.6
94.9
93.8
95.2
SVM
healthy
infection
(%)
OA (%)
Kappa
healthy
infection
(%)
OA (%)
Kappa
healthy
160
3
98.2
99.3
0.98
80
3
96.4
98.7
0.97
infection
0
292
100
0
142
100
(%)
100
98.9
100
97.9
Tables (8)
Table 1.
Technical Parameters of Cubert S185 Hyperspectral Imaging Sensor Mounted on UAV
Index
Parameters
Model
S185
Brand
Cubert
Dimension
Weight
0.47 Kg
Origin
Germany
Number of channels
125 bands
Spectral sampling
4 nm
Field of view
20°
Spectral wavebands
450–950 nm
Spectral resolution
8 nm at 532 nm
Type of datum
Panchromatic image and hyperspectral image
Sensor type
Silicon
Cooling
Passive air-cooled
Digitization
12 bit
Integration time
0.1–1000 ms
Table 2.
Details of the Vegetation Indices Used in the Studya
${R}$ represents band reflectance. The spectral sampling interval of S185 was 4 nm, and the bands centered in 680 or 800 did not existed in its band settings. Therefore, the reflectance of the band that is not in the band setting was calculated as an average of the reflectivity of the adjacent two bands (e.g., the reflectance of bands 680 or 800 in the S185 data were calculated as the average reflectance at 682 and 678 nm and at 802 and 798 nm, respectively).
Table 3.
Optimization of Vegetation Index Based on S185 Bandsa
Indicates the significant correlation in ${p}$-value ${\lt}\;{0.05}$.
Indicates the significant correlation in ${p}$-value ${\lt}\;{0.01}$.
Indicates the significant correlation in ${p}$-value ${\lt}\;{0.001}$.
Table 4.
Location and Scale Parameters of Sensitive Wavelet Features in Different Infestation Stages
Stages
Features
Range of Spectrum
Location (nm)
Scale
26 DAI
WF1
Green
474–546 (530)
0.76
WF2
Yellow
562–594 (574)
0.71
WF3
Red
574–682 (674)
0.74
WF4
Red-edge
706–710 (706)
0.67
WF5
Near infrared
754–762 (758)
0.67
42 DAI
WF1
Green
490–526 (510)
0.79
WF2
Red
618–678 (662)
0.82
WF3
Red-edge
694–726 (714)
0.85
WF4
Near infrared
762–774 (766)
0.82
WF5
Near infrared
786–794 (790)
0.82
Table 5.
Classification Accuracies of LDA and SVM Models Based on the Vegetation Index Features in 26 DAI Stagea
Methods
Training Data
Validation Data
LDA
healthy
infection
U (%)
OA (%)
Kappa
healthy
infection
U (%)
OA (%)
Kappa
healthy
134
39
77.4
85.7
0.69
62
12
83.8
86.7
0.70
infection
26
256
90.8
18
133
88.1
(%)
83.8
96.8
77.5
91.7
SVM
healthy
infection
U (%)
OA (%)
Kappa
healthy
infection
U (%)
OA (%)
Kappa
healthy
132
23
85.2
88.8
0.75
65
12
84.4
88.0
0.74
infection
28
272
90.1
15
133
89.9
(%)
82.5
92.2
81.3
91.7
For Table 5–8: P = producer’s accuracy, U = user’s accuracy, OA = overall accuracy.
Table 6.
Classification Accuracies of LDA and SVM Models Based on the Vegetation Index Features in 42 DAI Stage
Methods
Training Data
Validation Data
LDA
healthy
infection
U (%)
OA (%)
Kappa
healthy
infection
U (%)
OA (%)
Kappa
healthy
141
11
92.8
93.4
0.85
72
9
88.9
92.4
0.84
infection
19
284
93.8
8
136
94.4
(%)
88.1
96.3
90.0
93.8
SVM
healthy
infection
(%)
OA (%)
Kappa
healthy
infection
(%)
OA (%)
Kappa
healthy
153
15
91.7
95.2
0.95
80
6
93.0
97.3
0.94
infection
7
280
97.6
0
139
100
(%)
95.6
94.9
100
95.8
Table 7.
Classification Accuracies of LDA and SVM Models Based on the Continuous Wavelet Features in 26 DAI Stage
Methods
Training Data
Validation Data
LDA
healthy
infection
(%)
OA (%)
Kappa
healthy
infection
(%)
OA (%)
Kappa
healthy
125
21
85.6
87.7
0.72
64
14
82.1
86.7
0.71
infection
35
274
88.7
16
431
89.1
(%)
78.1
92.9
80.0
90.3
SVM
healthy
infection
(%)
OA (%)
Kappa
healthy
infection
(%)
OA (%)
Kappa
healthy
129
14
90.2
90.1
0.78
68
9
88.3
90.7
0.79
infection
31
281
90.1
12
136
91.9
(%)
80.6
95.3
85.0
93.8
Table 8.
Classification Accuracies of LDA and SVM Models Based on the Continuous Wavelet Features in 42 DAI Stage