农业工程学报
農業工程學報
농업공정학보
2014年
6期
116-123
,共8页
叶绿素%反向传播%回归分析%高光谱%4-scale模型
葉綠素%反嚮傳播%迴歸分析%高光譜%4-scale模型
협록소%반향전파%회귀분석%고광보%4-scale모형
chlorophyll%back propagation%regression analysis%hyperspectral%4-scale model
采用高光谱卫星数据进行玉米叶片和冠层尺度的叶绿素含量估算，对现代农业技术的发展有重要意义。首先，采用以α为倾斜度参数的双曲正切S型函数为基础的误差反向传播（back propagation，BP）算法前馈神经网络（hyperbolic tangent sigmoid function-back propagation，Htsf-BP）构建叶片尺度的叶绿素含量高光谱遥感估算模型；以几何光学辐射传输模型（4-scale 模型）为理论依据，对叶片和冠层尺度的光谱转化函数进行推导，实现Hyperion影像冠层尺度光谱到叶片光谱的转化，同时获取叶片尺度叶绿素含量估算结果；最后，结合叶面积指数（leaf area index，LAI）进行冠层尺度叶绿素含量估算。结果表明：当隐含层结点数为6时，Htsf-BP神经网络法对叶绿素的估算精度最高，验证精度达78.68%；在波长750与980 nm处，采用光谱尺度转化方程进行模拟的冠层光谱与实测冠层光谱间的相关系数R2值分别达到了0.784和0.706；实测叶片尺度叶绿素含量与模拟结果间的相关系数R2值达0.726。该方法可为高精度快速估算叶片和冠层尺度玉米叶绿素含量提供参考。
採用高光譜衛星數據進行玉米葉片和冠層呎度的葉綠素含量估算，對現代農業技術的髮展有重要意義。首先，採用以α為傾斜度參數的雙麯正切S型函數為基礎的誤差反嚮傳播（back propagation，BP）算法前饋神經網絡（hyperbolic tangent sigmoid function-back propagation，Htsf-BP）構建葉片呎度的葉綠素含量高光譜遙感估算模型；以幾何光學輻射傳輸模型（4-scale 模型）為理論依據，對葉片和冠層呎度的光譜轉化函數進行推導，實現Hyperion影像冠層呎度光譜到葉片光譜的轉化，同時穫取葉片呎度葉綠素含量估算結果；最後，結閤葉麵積指數（leaf area index，LAI）進行冠層呎度葉綠素含量估算。結果錶明：噹隱含層結點數為6時，Htsf-BP神經網絡法對葉綠素的估算精度最高，驗證精度達78.68%；在波長750與980 nm處，採用光譜呎度轉化方程進行模擬的冠層光譜與實測冠層光譜間的相關繫數R2值分彆達到瞭0.784和0.706；實測葉片呎度葉綠素含量與模擬結果間的相關繫數R2值達0.726。該方法可為高精度快速估算葉片和冠層呎度玉米葉綠素含量提供參攷。
채용고광보위성수거진행옥미협편화관층척도적협록소함량고산，대현대농업기술적발전유중요의의。수선，채용이α위경사도삼수적쌍곡정절S형함수위기출적오차반향전파（back propagation，BP）산법전궤신경망락（hyperbolic tangent sigmoid function-back propagation，Htsf-BP）구건협편척도적협록소함량고광보요감고산모형；이궤하광학복사전수모형（4-scale 모형）위이론의거，대협편화관층척도적광보전화함수진행추도，실현Hyperion영상관층척도광보도협편광보적전화，동시획취협편척도협록소함량고산결과；최후，결합협면적지수（leaf area index，LAI）진행관층척도협록소함량고산。결과표명：당은함층결점수위6시，Htsf-BP신경망락법대협록소적고산정도최고，험증정도체78.68%；재파장750여980 nm처，채용광보척도전화방정진행모의적관층광보여실측관층광보간적상관계수R2치분별체도료0.784화0.706；실측협편척도협록소함량여모의결과간적상관계수R2치체0.726。해방법가위고정도쾌속고산협편화관층척도옥미협록소함량제공삼고。
The chlorophyll content estimation of corn leaf and canopy scale using hyperspectral satellite data has important significance to the development of modern agriculture technology. First of all, 60 separate 30 m by 30 m sample plots were set up randomly in a research area, and the excellent, good, and bad plants in each sample plot were selected as sample plants based on corn growth conditions. The plant canopy was divided into three layers and 5-10 pieces of fresh leaves samples were collected one layer at a time to test chlorophyll and spectrum. Three sample points (each point between 10 and 15 m apart) in each sample plot were chosen to test the canopy spectrum, and the average of three replicates was taken as the canopy spectral values. The LAI - 2000 canopy analyzer was used to determine the corn leaf area index (LAI). Secondly, the first-order differential (FD (Ref)) and curve remove (CR (Ref)) methods were used to process original reflectivity spectrum (Ref), and 5 Ref variables, 1 CR (Ref) variables, and 7 FD (Ref) variables that had a higher correlation with leaf chlorophyll content, simple ratio index (SR), and normalized differential vegetation index (NDVI) were chosen to construct a multiple stepwise regression model. At the 95%significance level, 13 variables were eliminated, and a multiple linear regression model that only contains two variables was established. Variables of R890 and R1070 that were kept by multiple linear regression models were used as the input parameters of Back Propagation neural networks based on a hyperbolic tangent sigmoid function with slope parameters of a (Htsf-BP). So, a remote sensing estimation model by hyperspectral of chlorophyll of leaf scale was built, and when the number of neurons was six, the overall precision of Htsf-BP neural network was the best, with the fitting precision of 88.73%, verification accuracy of 78.68%, and validation RMSE of 0.0704. Sensor observation information, the structure parameters of the plant, background spectrum, and leaf spectral were put into a 4-scale model to simulate canopy reflectance on the different conditions. According to the data sets simulation of the 4-scale model, the fitting relationship between PT, PG and LAI functions could be obtained, and the solving results of factor M and b could be obtained at the same time. To move forward a single step, the conversion of the Hyperion image spectrum of the canopy and the leaf spectrum could be implemented, and the chlorophyll content estimation results of the Leaf blade scale could be achieved. At last, the chlorophyll content estimation of the canopy scale combining with the leaf area index could be accomplished. The results have showed that at the wavelength of 750 and 980 nm, the correlation coefficient R2 value between canopy spectra simulated by spectral dimension transformation equation and the measured canopy spectra reached 0.784 and 0.706 respectively. The correlation coefficient R2 value between measured leaf dimension chlorophyll content and the simulation results was 0.726. This method has provided a reference for the high precision fast estimation of leaf and canopy scale corn chlorophyll content.
