高光谱图像和叶绿素含量的水稻纹枯病早期检测识别
|
摘要
基于高光谱成像技术和化学计量方法,实现了对水稻纹枯病病害的早期检测识别。以幼苗时期的水稻植株为研究对象,对其进行纹枯病病菌侵染,获得染病植株,采集358～1 021 nm波段范围的高光谱图像,三次实验共240个样本,包括染病植株120个样本和健康植株120个样本。根据高光谱图像的光谱维,对染病水稻叶片和健康水稻叶片提取感兴趣区域(ROI),利用感兴趣区域的光谱数据,对其进行Savitzky-Golay(SG)平滑、 Savitzky-Golay(SG)一阶求导、 Savitzky-Golay(SG)二阶求导、变量标准化(SNV)和多元散射校正(MSC)预处理,建立线性判别分析(LDA)和支持向量机(SVM)分类模型,结果表明:采用SG二阶求导预处理后的线性判别分析(LDA)模型取得了较好的性能,正确识别率在建模集达98.3%,在预测集达95%;利用载荷系数法(x-loading weights,x-LW)对原始光谱和5种预处理的光谱数据进行特征波长提取,然后根据选取的特征波长建立线性判别分析(LDA)和支持向量机(SVM)分类模型,其中采用SG二阶求导预处理后提取的12个特征波长的线性判别分析(LDA)模型取得了较好的性能,其正确识别率在建模集达97.8%,在预测集达95%,而且基于载荷系数法建立的模型性能与全波段相当,可以通过载荷系数法减少数据量对水稻纹枯病病害进行识别;根据高光谱图像的图像维,研究了基于图像主成分分析、基于概率滤波和基于二阶概率滤波的图像特征提取方法,利用提取的特征变量建立反向传播神经网络(BPNN)和支持向量机(SVM)分类模型,其中基于图像主成分分析的反向传播神经网络(BPNN)模型取得了较好的性能,建模集准确识别率达90.6%,预测集的准确识别率达83.3%;根据高光谱图像光谱维和图像维的最优模型,特将叶绿素含量作为建模的另一个特征,分别与光谱特征、图像特征组合,建立反向传播神经网络(BPNN)和线性判别分析(LDA)模型,提出基于光谱特征加叶绿素含量、图像特征加叶绿素含量和光谱、图像特征加叶绿素含量三种组合方式,其中,光谱特征和图像特征分别与叶绿素组合的方式比之前单独的光谱和图像特征建模性能都有所提升,而且三种组合方式中光谱特征加叶绿素含量的反向传播神经网络(BPNN)建模方式取得本研究所有建模方式中较优的性能,其准确识别率在建模集达100%,在预测集达96.7%。以上研究表明,基于高光谱图像和叶绿素含量对水稻纹枯病病害进行早期识别是可行的,为水稻病害的早期识别提供了一种新方法。
Hyperspectral imaging combined with chemometrics was successfully proposed to identify the rice sheath blight disease. First, infected rice plants with rice sheath blight in the seedling period to get the infected rice plants, then used the hyperspectral imaging system to acquire the hyperspectral imagines in the spectral range of 358～1 021 nm, finally selected 240 samples of all hyperspectral imagines to analyze, including 120 healthy samples and 120 infected samples. According to the spectral dimension of hyperspectral image, extracted the region of interest(ROI) of healthy and infected rice leaves, pretreated the spectral data of the region of interest with pretreatments including SG smoothing, SG-1 D, SG-2 D, SNV and MSC, then established the linear discriminant analysis(LDA) and support vector machine(SVM) classification models. The result showed that the linear discriminant analysis(LDA) model with SG-2 D pretreatment achieved the better performance, with the correct recognition rate of the modeling set being 98.3% and the correct recognition rate of the prediction set being 95%. After five kinds of pretreatments, extracted the feature wavelengths with the method of x-loading weights, then established the linear discriminant analysis(LDA) and support vector machine(SVM) classification models based on feature wavelengths. The result showed that the linear discriminant analysis(LDA) model with SG-2 D pretreatment achieved the better performance, with the correct recognition rate being 97.8% in the modeling set and 95% in the prediction set. Moreover, the model performance based on x-loading weights was equivalent to that of the whole band. So, it can be used to identify the rice sheath blight disease with x-loading weights. According to the image dimension of hyperspectral image, the principal component analysis, probabilistic filtering and second-order probabilistic filtering were proposed in this paper, then established the back propagation neural network(BPNN) and support vector machine(SVM) classification models. The result showed that the BPNN based on image principal component analysis achieved the better performance, with the correct recognition rate being 90.6% in the modeling set and 93.3% in the prediction set. According to the spectral and image dimension of hyperspectral image, the chlorophyll content was proposed to be another feature of disease recognition, which was combined with spectral characteristics and image features to build models to compare the performance of each model. Then established the back propagation neural network(BPNN) and linear discriminant analysis(LDA) classification models. The spectral features combining with chlorophyll content, image features combining with chlorophyll content and spectral, image features combining with chlorophyll content were proposed. The performance of spectral, image features combining with chlorophyll respectively were both better than that using the spectral and image features alone. BPNN based on spectral features combining with chlorophyll content achieved the better performance, with the correct recognition rate being 100% in the modeling set and 96.7% in the prediction set, also, this model achieved the best performance compared with all models in this paper. The overall results indicated that hyperspectral imaging technology with chlorophyll content can accurately identify the early rice sheath blight disease and provide a new method for early detection of rice disease.
Hyperspectral imaging combined with chemometrics was successfully proposed to identify the rice sheath blight disease. First, infected rice plants with rice sheath blight in the seedling period to get the infected rice plants, then used the hyperspectral imaging system to acquire the hyperspectral imagines in the spectral range of 358～1 021 nm, finally selected 240 samples of all hyperspectral imagines to analyze, including 120 healthy samples and 120 infected samples. According to the spectral dimension of hyperspectral image, extracted the region of interest(ROI) of healthy and infected rice leaves, pretreated the spectral data of the region of interest with pretreatments including SG smoothing, SG-1 D, SG-2 D, SNV and MSC, then established the linear discriminant analysis(LDA) and support vector machine(SVM) classification models. The result showed that the linear discriminant analysis(LDA) model with SG-2 D pretreatment achieved the better performance, with the correct recognition rate of the modeling set being 98.3% and the correct recognition rate of the prediction set being 95%. After five kinds of pretreatments, extracted the feature wavelengths with the method of x-loading weights, then established the linear discriminant analysis(LDA) and support vector machine(SVM) classification models based on feature wavelengths. The result showed that the linear discriminant analysis(LDA) model with SG-2 D pretreatment achieved the better performance, with the correct recognition rate being 97.8% in the modeling set and 95% in the prediction set. Moreover, the model performance based on x-loading weights was equivalent to that of the whole band. So, it can be used to identify the rice sheath blight disease with x-loading weights. According to the image dimension of hyperspectral image, the principal component analysis, probabilistic filtering and second-order probabilistic filtering were proposed in this paper, then established the back propagation neural network(BPNN) and support vector machine(SVM) classification models. The result showed that the BPNN based on image principal component analysis achieved the better performance, with the correct recognition rate being 90.6% in the modeling set and 93.3% in the prediction set. According to the spectral and image dimension of hyperspectral image, the chlorophyll content was proposed to be another feature of disease recognition, which was combined with spectral characteristics and image features to build models to compare the performance of each model. Then established the back propagation neural network(BPNN) and linear discriminant analysis(LDA) classification models. The spectral features combining with chlorophyll content, image features combining with chlorophyll content and spectral, image features combining with chlorophyll content were proposed. The performance of spectral, image features combining with chlorophyll respectively were both better than that using the spectral and image features alone. BPNN based on spectral features combining with chlorophyll content achieved the better performance, with the correct recognition rate being 100% in the modeling set and 96.7% in the prediction set, also, this model achieved the best performance compared with all models in this paper. The overall results indicated that hyperspectral imaging technology with chlorophyll content can accurately identify the early rice sheath blight disease and provide a new method for early detection of rice disease.
引文
[1] WANG Yue-ming,JIA Jian-xin,HE Zhi-ping,et al(王跃明,贾建鑫,何志平,等).Journal of Remote Sensing(遥感学报),2016,20(5):850.
[2] WANG Ren-hong,SONG Xiao-yu,LI Zhen-hai,et al(王仁红,宋晓宇,李振海,等).Trancations of the Chinese Society of Agricatural Engineering(农业工程学报),2014,30(19):191.
[3] HE Ru-yan,QIAO Xiao-jun,JIANG Jin-bao,et al(何汝艳,乔小军,蒋金豹,等).Trancations of the Chinese Society of Agricatural Engineering(农业工程学报),2015,31(2):141.
[4] LIU Ke,ZHOU Qing-bo,WU Wen-bin,et al(刘轲,周清波,吴文斌,等).Trancations of the Chinese Society of Agricatural Engineering(农业工程学报),2016,32(3):155.
[5] YANG Fu-qin,FENG Hai-kuan,LI Zhen-hai,et al(杨福芹,冯海宽,李振海,等).Trancations of the Chinese Society of Agricatural Engineering(农业工程学报),2016,32(3):163.
[6] Paolo Menesatti,Angelo Zanella,Stefano D’Andrea,et al.Food and Bioprocess Technology,2009,2(3).
[7] DENG Xiao-lei,LI Min-zan,ZHENG Li-hua(邓小蕾,李民赞,郑立华,等).Trancations of the Chinese Society of Agricatural Engineering(农业工程学报),2014,30(14):140.
[8] FENG Hai-kuan,YANG Fu-qin,LI Zhen-hai,et al(冯海宽,杨福芹,李振海,等).Trancations of the Chinese Society of Agricatural Engineering(农业工程学报),2016,32(7):173.
[9] ZHANG Shuai-tang,WANG Zi-yan,ZOU Xiu-guo,et al(张帅唐,王紫烟,邹修国,等).Trancations of the Chinese Society of Agricatural Engineering(农业工程学报),2017,33(22):200.
[10] ZHANG Song-lan(张松兰).Journal of Jiangsu University of Technology(江苏理工学院学报),2016,22(2):14.
[11] YU Lei,SHI Feng,WANG Hui,et al(郁磊,史峰,王辉,等).Analyses of 30 Cases of Artificial Intelligence Algorithms in MATLAB(MATLAB智能算法30个案例分析).Beijing:Beijing University of Aeronautics and Astronautics Press(北京:北京航空航天出版社),2015.27.
[2] WANG Ren-hong,SONG Xiao-yu,LI Zhen-hai,et al(王仁红,宋晓宇,李振海,等).Trancations of the Chinese Society of Agricatural Engineering(农业工程学报),2014,30(19):191.
[3] HE Ru-yan,QIAO Xiao-jun,JIANG Jin-bao,et al(何汝艳,乔小军,蒋金豹,等).Trancations of the Chinese Society of Agricatural Engineering(农业工程学报),2015,31(2):141.
[4] LIU Ke,ZHOU Qing-bo,WU Wen-bin,et al(刘轲,周清波,吴文斌,等).Trancations of the Chinese Society of Agricatural Engineering(农业工程学报),2016,32(3):155.
[5] YANG Fu-qin,FENG Hai-kuan,LI Zhen-hai,et al(杨福芹,冯海宽,李振海,等).Trancations of the Chinese Society of Agricatural Engineering(农业工程学报),2016,32(3):163.
[6] Paolo Menesatti,Angelo Zanella,Stefano D’Andrea,et al.Food and Bioprocess Technology,2009,2(3).
[7] DENG Xiao-lei,LI Min-zan,ZHENG Li-hua(邓小蕾,李民赞,郑立华,等).Trancations of the Chinese Society of Agricatural Engineering(农业工程学报),2014,30(14):140.
[8] FENG Hai-kuan,YANG Fu-qin,LI Zhen-hai,et al(冯海宽,杨福芹,李振海,等).Trancations of the Chinese Society of Agricatural Engineering(农业工程学报),2016,32(7):173.
[9] ZHANG Shuai-tang,WANG Zi-yan,ZOU Xiu-guo,et al(张帅唐,王紫烟,邹修国,等).Trancations of the Chinese Society of Agricatural Engineering(农业工程学报),2017,33(22):200.
[10] ZHANG Song-lan(张松兰).Journal of Jiangsu University of Technology(江苏理工学院学报),2016,22(2):14.
[11] YU Lei,SHI Feng,WANG Hui,et al(郁磊,史峰,王辉,等).Analyses of 30 Cases of Artificial Intelligence Algorithms in MATLAB(MATLAB智能算法30个案例分析).Beijing:Beijing University of Aeronautics and Astronautics Press(北京:北京航空航天出版社),2015.27.