
基于无人机多光谱影像的棉花黄萎病监测
宋勇, 陈兵, 王琼, 王刚, 王静, 刘海建, 郑登科, 李金霞, 陈子杰, 孙乐鑫
基于无人机多光谱影像的棉花黄萎病监测
Monitoring of cotton Verticillium wilt based on unmanned aerial vehicle multispectral images
【目的】通过无人机多光谱影像监测棉田黄萎病发病情况,为棉花黄萎病的精准防控提供理论指导。【方法】通过分析黄萎病发病棉田冠层的光谱特征,筛选无人机多光谱影像识别棉花黄萎病的最佳植被指数、最佳波段组合及最佳时相,并基于筛选的最佳时相建立黄萎病不同发病程度的棉田影像图,利用平行六面体法、最大似然法及支持向量机径向基函数分类法对影像图进行分类对比和精度评价。【结果】结果表明,在710~760 nm波段,不同发病程度的棉株冠层光谱反射率均随着波长的增加明显增加;在760~950 nm波段,棉株冠层光谱反射率随着黄萎病的加重明显减小。随着黄萎病加重发生,棉株的叶片叶绿素含量、地上部鲜物质质量、地上部干物质质量、植株含水量以及叶面积指数均降低。无人机多光谱遥感识别棉花黄萎病的最佳植被指数和最佳波段组合分别是差值植被指数(difference vegetation index, DVI)和B3-5-8(对应的波长分别为550 nm、656 nm和800 nm)。8月中下旬为无人机多光谱遥感识别棉花黄萎病发生程度的最佳时相。支持向量机径向基函数分类法结合最佳波段组合B3-5-8与DVI综合影像对棉田黄萎病病情的分类精度最高(分类精度为96.64%,Kappa系数为95.61%)。棉田黄萎病发病程度的分类结果与棉株冠层光谱反射率和棉株农学参数的变化相对应,并与实地调查结果一致。【结论】利用支持向量机径向基函数分类法、最佳波段组合B3-5-8与DVI综合影像对棉田黄萎病发病情况进行分类是可行的,研究结果可为利用遥感技术监测作物类似病虫害提供理论依据。
[Objective] The unmanned aerial vehicle (UAV) multi-spectral remote sensing technology was used to monitor the severity of cotton Verticillium wilt, which will provide theoretical guidance for the precise prevention and control of cotton Verticillium wilt. [Method] By analyzing the spectral characteristics of canopy in cotton field affected by Verticillium wilt, the best vegetation index, the best wavelength combination and the best time phase for multi-spectral identification of Verticillium wilt by UAV were selected. The images of cotton fields differing in the severity of Verticillium wilt were established based on the optimal time phase. Parallelepiped method, maximum likelihood method and support vector machine radial basis function classification method were used to classify and evaluate the accuracy of the images. [Result] The results showed that the canopy spectral reflectance of cotton plants differing with occurrence of Verticillium wilt increased obviously with the increase of wavelength at 710-760 nm, the spectral reflectance of cotton canopy decreased obviously with the aggravation of Verticillium wilt at the wavelength of 760-950 nm. With the aggravation of Verticillium wilt, the chlorophyll content of leaves, fresh mass of aerial tissue per plant, dry mass of aerial tissue per plant, plant water content and leaf area index of cotton plants were all decreased. The best vegetation index and best band combination for UAV multi-spectral remote sensing to identify cotton Verticillium wilt were difference vegetation index (DVI) and B3-5-8 (the corresponding wavelengths are 550 nm, 656 nm and 800 nm). Mid to late August was the best time to identify the occurrence degree of cotton Verticillium wilt by UAV multi-spectral remote sensing. Support vector machine radial basis function classification method, best band combination B3-5-8 and DVI integrated image had the highest classification accuracy for Verticillium wilt in cotton field (classification accuracy was 96.64%, and Kappa coefficient was 95.61%). The classification results of Verticillium wilt severity in cotton fields corresponded to the changes in canopy spectral reflectance and agronomic parameters of cotton plants, and were consistent with the field investigation results. [Conclusion] It is feasible to classify the severity of Verticillium wilt in cotton field by using support vector machine radial basis function classification method, optimal band combination B3-5-8 and DVI integrated images. The results can provide theoretical ground for monitoring similar crop pests and diseases by remote sensing technology.
棉花 / 黄萎病 / 无人机 / 多光谱影像 / 分类方法 {{custom_keyword}} /
cotton / Verticillium wilt / unmanned aerial vehicle / multispectral image / classification method {{custom_keyword}} /
表1 棉花黄萎病病害严重度分级标准Table 1 Classification standard of cotton Verticillium wilt disease severity |
病害等级 Disease severity level | 病害严重度 Disease severity | 病情指数 Disease index (DI) |
---|---|---|
b0 | 健康 Health | 0 |
b1 | 轻度Slight | 0<DI≤25 |
b2 | 中度Moderate | 25<DI≤50 |
b3 | 重度Serious | 50<DI≤75 |
b4 | 极严重Critical | 75<DI≤100 |
表2 本研究中选取的15种植被指数Table 2 Fifteen vegetation indexes selected in this study |
植被指数 Vegetation index | 公式 Formula | 参考文献 Reference |
---|---|---|
归一化差值植被指数 Normalized difference vegetation index (NDVI) | [20] | |
比值植被指数 Ratio vegetation index (RVI) | [20] | |
差值植被指数 Difference vegetation index (DVI) | [21] | |
重归一化差值植被指数 Re-normalized difference vegetation index (RDVI) | [21] | |
绿波段归一化差值植被指数 Green band normalized difference vegetation index (GNDVI) | [21] | |
红边归一化差值植被指数 Red edge normalized difference vegetation index (RENDVI) | [22] | |
归一化差异绿度指数 Normalized difference greenness index (NDGI) | [22] | |
三角植被指数 Triangle vegetation index (TVI) | [23] | |
土壤调节植被指数 Soil adjusted vegetation index (SAVI) | [23] | |
优化土壤调节植被指数 Optimize soil adjusted vegetation index (OSAVI) | [24] | |
调整型土壤调节植被指数 Modified soil-adjusted vegetation index (MSAVI) | [24] | |
花青素反射指数 Anthocyanin reflex index (ARI) | [25] | |
增强型植被指数 Enhanced vegetation index (EVI) | [25] | |
归一化差异水体指数 Normalized difference water index (NDWI) | [25] | |
水波段指数 Water band index (WBI) | [25] |
注:R550、R656、R710、R760、R800、R900和R950分别表示波长550 nm、656 nm、710 nm、760 nm、800 nm、900 nm、950 nm处的棉花冠层光谱反射率。 | |
Note: R550, R656, R710, R760, R800, R900 and R950 represent the spectral reflectance value of cotton canopy at the wavelengths of 550 nm, 656 nm, 710 nm, 760 nm, 800 nm, 900 nm, and 950 nm, respectively. |
表3 黄萎病不同发生程度棉田棉株农学参数变化Table 3 Changes of agronomy parameters of cotton in fields differing in the severity of Verticillium wilt |
病害严重度等级 Disease severity level | 叶绿素含量 Chlorophyll content/(mg·dm-2) | 单株地上部鲜物质质量Fresh mass of aerial tissue per plant/g | 单株地上部干物质质量 Dry mass of aerial tissue per plant/g | 植株含水量 Plant water content/% | 叶面积指数 Leaf area index |
---|---|---|---|---|---|
b0 | 7.95±0.02 a | 323.63±29.69 a | 77.70±13.36 a | 76.02±1.01 a | 4.79±0.35 a |
b1 | 6.05±0.02 ab | 251.74±19.81 ab | 62.69±19.67 ab | 74.93±1.02 a | 4.09±0.28 ab |
b2 | 4.85±0.01 b | 193.20±16.57 b | 51.74±14.14 b | 69.26±2.98 ab | 2.38±0.23 b |
b3 | 3.30±0.01 c | 111.55±30.54 bc | 39.08±16.02 c | 65.43±1.21 b | 1.44±0.21 c |
b4 | 2.12±0.01 cd | 56.17±16.96 c | 21.12±6.69 cd | 61.07±1.18 bc | 0.71±0.25 cd |
注:数据为平均值±标准差。同列不同小写字母表示在0.05水平差异显著。 | |
Note: Data are mean ± standard deviation. Different lowercase letters in the same column indicate significant difference at the 0.05 probability level. |
表4 棉花黄萎病病情指数与15个植被指数的相关分析Table 4 Correlation analysis of 15 vegetation indexes and disease index of cotton Verticillium wilt |
植被指数 Vegetation index | 相关系数 Correlation coefficient | 植被指数 Vegetation index | 相关系数 Correlation coefficient | |
---|---|---|---|---|
DVI | -0.86** | RENDVI | -0.48** | |
TVI | -0.73** | NDVI | -0.37** | |
RDVI | -0.71** | EVI | -0.36** | |
SAVI | -0.70** | NDGI | 0.30** | |
MSAVI | -0.68** | RVI | -0.27** | |
GNDVI | -0.67** | WBI | 0.16** | |
ARI | -0.56** | NDWI | 0.12* | |
OSAVI | -0.55** |
注:**和*分别表示极显著相关(P<0.01)和显著相关(P<0.05)。 | |
Note: ** and * indicate extremely significant correlation (P < 0.01) and significant correlation (P < 0.05), respectively. |
表5 12个波段下棉花冠层灰度值标准差Table 5 Standard deviations of gray value of cotton canopy under 12 bands |
波段 Band | 标准差 Standard deviation | 排序 Rank | 波段 Band | 标准差 Standard deviation | 排序 Rank | |
---|---|---|---|---|---|---|
B1 | 10.27 | 9 | B7 | 18.73 | 3 | |
B2 | 12.71 | 5 | B8 | 11.41 | 6 | |
B3 | 14.26 | 4 | B9 | 7.66 | 11 | |
B4 | 11.16 | 8 | B10 | 8.56 | 10 | |
B5 | 21.77 | 2 | B11 | 11.22 | 7 | |
B6 | 30.54 | 1 | B12 | 7.56 | 12 |
表6 OIF值前15位的波段组合Table 6 The top 15 band combinations based on OIF value |
序号 Number | 波段组合 Band combination | OIF值 OIF value | 序号 Number | 波段组合 Band combination | OIF值 OIF value | |
---|---|---|---|---|---|---|
1 | B3、B5、B8 | 153.44 | 9 | B1、B6、B10 | 81.88 | |
2 | B4、B6、B8 | 132.26 | 10 | B1、B7、B8 | 80.98 | |
3 | B4、B6、B9 | 128.28 | 11 | B1、B6、B8 | 79.61 | |
4 | B4、B6、B10 | 109.57 | 12 | B1、B6、B9 | 79.21 | |
5 | B2、B3、B8 | 97.48 | 13 | B4、B5、B9 | 73.56 | |
6 | B4、B5、B8 | 91.76 | 14 | B3、B6、B9 | 70.73 | |
7 | B3、B6、B8 | 89.27 | 15 | B2、B8、B11 | 67.83 | |
8 | B3、B5、B9 | 83.71 |
图4 基于不同光谱影像的棉花黄萎病发生程度分类A、D、G分别为基于原始波段影像、最佳波段组合(B3-5-8)影像、最佳波段组合与DVI综合影像的平行六面体分类结果;B、E、H分别为基于原始波段影像、最佳波段组合影像、最佳波段组合与DVI综合影像的最大似然分类结果;C、F、I分别为基于原始波段影像、最佳波段组合影像、最佳波段组合与DVI综合影像的支持向量机径向基函数分类结果。Fig. 4 Classification of cotton Verticillium wilt severity degree based on different spectral images A, D and G are the parallelepiped classification results based on the original band images, the best band combination (B3-5-8 )images, the best combined band images and the DVI integrated images, respectively. B, E and H are the maximum likelihood classification results based on the original band images, the best band combination images, the best combined band images and the DVI integrated images, respectively. C, F and I are the support vector machine radial basis function classification results based on the original band images, the best band combination images, the best combined band images and DVI integrated images, respectively. |
表7 不同分类模型的影像分类精度参数Table 7 Classification accuracy parameters of images using different classification models |
影像类型 Image type | 平行六面体分类法 Parallelepiped classification | 最大似然分类法 Maximum likehood classification | 支持向量机径向基函数 分类法 Support vector machine radial basis function classification | |||||
---|---|---|---|---|---|---|---|---|
分类精度 Classification accuracy/% | Kappa系数 Kappa coefficient/% | 分类精度 Classification accuracy/% | Kappa系数 Kappa coefficient/% | 分类精度 Classification accuracy/% | Kappa系数 Kappa coefficient% | |||
原始波段影像 Original band image | 90.24 | 87.39 | 88.51 | 85.26 | 61.65 | 53.51 | ||
最佳波段组合影像 Best band combination image | 90.47 | 87.52 | 94.58 | 93.03 | 88.99 | 85.62 | ||
最佳波段组合与DVI综合影像 Best combined band images and DVI integrated image | 91.46 | 88.89 | 94.64 | 93.12 | 96.64 | 95.61 |
[1] |
朱荷琴, 李志芳, 冯自力, 等. 我国棉花黄萎病研究十年回顾及展望[J/OL]. 棉花学报, 2017, 29(S1): 37-50[2022-09-20]. https://doi.org/10.11963/1002-7807.zhqzhq.20170825.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[2] |
王姣, 李志沛, 张立福, 等. 基于棉花黄萎病多“症状”特征的植被指数构建及病情遥感监测研究[J/OL]. 地理与地理信息科学, 2019, 35(5): 46-51[2022-09-20]. https://doi.org/10.3969/j.issn.1672-0504.2019.05.008.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[3] |
陈兵, 李少昆, 王克如, 等. 基于TM影像光谱指数的棉花病害严重度估测[J]. 红外与毫米波学报, 2011, 30(5): 451-457.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[4] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[5] |
陈巧玲, 范燕敏, 武红旗, 等. 棉花黄萎病冠层光谱特征分析及其严重度估测模型的建立[J/OL]. 新疆农业大学学报, 2020, 43(4): 261-269[2022-09-20]. https://doi.org/10.3969/j.issn.1007-8614.2020.04.004.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[6] |
陈兵, 王琼, 肖春华, 等. 9种胁迫条件下棉花叶片光谱特征及其识别研究[J/OL]. 西北农业学报, 2015, 24(10): 64-73[2022-09-20]. https://doi:10.7606/j.issn.1004-1389.2015.10.010.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[7] |
陈兵, 李少昆, 王克如, 等. 棉花黄萎病病叶光谱特征与病情严重度的估测[J]. 中国农业科学, 2007, 40(12): 2709-2715.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[8] |
竞霞, 王纪华, 宋晓宇, 等. 棉花黄萎病病情严重度的连续统去除估测法[J/OL]. 农业工程学报, 2010, 26(1): 193-198[2022-09-20]. https://doi.org/10.3969/j.issn.1002-6819.2010.01.034.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[9] |
陈兵. 基于多平台棉花黄萎病的遥感监测研究[D]. 石河子: 石河子大学, 2010.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[10] |
竞霞, 黄文江, 琚存勇, 等. 基于PLS算法的棉花黄萎病高空间分辨率遥感监测[J/OL]. 农业工程学报, 2010, 26(8): 229-235[2022-09-20]. https://doi.org/10.3969/j.issn.1002-6819.2010.08.039.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[11] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[12] |
陈兵. 棉花黄萎病高光谱特征及其光谱诊断模型研究[D]. 石河子: 石河子大学, 2007.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[13] |
石志钰, 裴雅琨, 朱玉涛, 等. 基于数字图像棉花黄萎病诊断与防治的管理远程监测[J/OL]. 新疆农业科学, 2020, 57(6): 1166-1174[2022-09-20]. https://doi.org/10.6048/j.issn.1001-4330.2020.06.022.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[14] |
高攀. 基于图像识别的棉花黄萎病实时诊断方法研究[D]. 石河子: 石河子大学, 2019.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[15] |
陈兵, 王克如, 李少昆, 等. 棉花黄萎病疑似病田的卫星遥感监测——以TM卫星影像为例[J/OL]. 作物学报, 2012, 38(1): 129-139[2022-09-20]. https://doi.org/10.3724/SP.J.1006.2012.00129.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[16] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[17] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[18] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[19] |
中华人民共和国农业部. 棉花抗病虫性评价技术规范第5部分: 黄萎病 GB/T 22101.5-2009[S]. 北京: 中国标准出版社, 2009.
Ministry of Agriculture of the People's Republic of China. Technical specification for evaluating resistance of cotton to disease and insect pests-Part 5: Verticillium wilt GB/T 22101.5-2009[S]. Beijing: China Standards Press, 2009.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[20] |
陈兵, 李少昆, 王克如, 等. 病害胁迫下棉花叶片色素含量高光谱遥感估测研究[J/OL]. 光谱学与光谱分析, 2010, 30(2): 421-425[2022-09-20]. https://doi.org/10.3964/j.issn.1000-0593(2010)02-0421-05.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[21] |
田庆久, 闵祥军. 植被指数研究进展[J/OL]. 地球科学进展, 1998(4):10-16[2022-09-20]. https://doi.org/10.3321/j.issn:1001-8166.1998.04.002.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[22] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[23] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[24] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[25] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[26] |
赵庆展, 刘伟, 尹小君, 等. 基于无人机多光谱影像特征的最佳波段组合研究[J/OL]. 农业机械学报, 2016, 47(3): 242-248, 291[2022-09-20]. https://doi.org/10.6041/j.issn.1000-1298.2016.03.034.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[27] |
洪帅, 张泽, 张立福, 等. 滴灌棉花不同生育时期冠层叶片叶绿素含量的高光谱估测模型[J/OL]. 棉花学报, 2019, 31(2): 138-146[2022-09-20]. https://doi.org/10.11963/1002-7807.hslx.20190319.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[28] |
陈兵, 王方永, 韩焕勇, 等. 基于光谱红边参数的棉花黄萎病叶片氮素含量诊断研究[J/OL]. 棉花学报, 2013, 25(3): 254-261[2022-09-20]. https://doi.org/10.11963/cs130310.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[29] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[30] |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
[31] |
朱耀辉. 基于无人机成像高光谱的冬小麦全蚀病等级监测[D]. 郑州: 河南农业大学, 2018.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
{{custom_ref.label}} |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
/
〈 |
|
〉 |