Hyperspectral Inversion Of Relative Chlorophyll Content In Cotton Canopy Leaves

Cotton is China’s main crop,the most important raw material for textile and fine chemical industries,and has an unshakable and important role in China’s national economy.In recent years,the problems affecting cotton growth status and production capacity have become a hot topic in the field of agronomy.Chlorophyll is the most important pigment in the growth and development of cotton,which can indirectly reflect the degree of photosynthesis of cotton,nitrogen utilization and the ability to exchange energy with the outside world.Therefore,it is of great research significance to monitor the chlorophyll of cotton.Hyperspectral technology can monitor the physiological and biochemical characteristics of crops in real time,without loss and speed,analyze the correlation between the hyperspectral data of the feature and the relative content of chlorophyll in the leaves of the cotton canopy through statistical methods,and construct an inversion model between the two,which can realize the monitoring of the growth and nutritional status of cotton,so as to provide certain theoretical support for the health diagnosis and formula fertilization of cotton.In this study,the data collected by the Agricultural Teaching and Research Practice Base of Tarim University in Aral City,Xinjiang Province were analyzed,and the hyperspectral data of cotton canopy leaves were first reduced in dimensionality,and then SVM,random forest and BP neural network models that estimated the relative content of chlorophyll in leaves were established based on the dimensionality reduction data,and various models were compared and evaluated.The main work and research results are as follows:(1)The relative original spectral parameters of cotton canopy leaves were statistically calculated,and it was found that the overall spectral reflectance of cotton canopy leaves at the seedling stage,bud stage and flowering boll stage was the same,and the spectral reflectance trend of each period in the range of 400 nm-500 nm was flat and the difference was small,and a strong reflection peak appeared in the middle of 500 nm-600 nm,and an absorption valley was formed in the middle position of 650 nm-700 nm The reflectivity rises rapidly between the680nm-760nm bands,and a high reflectance stabilization platform with significant continuity is formed between the 780nm-1000nm bands.(2)The original spectrum and red edge of cotton in three growth periods were analyzed,as well as the original spectrum and red edge change law of cotton under different chlorophyll content.Studies have shown that with the advancement of the growth period,the spectral reflectance of cotton canopy leaves will gradually increase in the long-wave band and near-infrared band of visible light.During the boll stage,both the red edge area and the red edge amplitude reach the maximum value.Cotton canopy leaves with different chlorophyll content also show different laws,in the short-wave stage of the visible light band,accompanied by the increase of the relative content of chlorophyll,the spectral reflectance of the cotton canopy leaves has a less obvious trend of gradual decrease,in the long-wave and near-infrared bands of visible light,accompanied by the increase of the relative content of chlorophyll,the spectral reflectivity of the cotton canopy leaves shows an upward trend,and there is the same"redshift"phenomenon as green plants such as rice and corn.(3)Aiming at the problem of high spectral data dimension of cotton canopy leaves,the characteristic bands of the first-order differential spectra of cotton canopy leaves were extracted by continuous projection algorithm with good pretreatment effect,among which the number of characteristic bands extracted at the seedling stage was 9,the number of bud stages was 7,and the number of flower bell stages(20)was 4,and the number of feature bands(20)screened out only occupied 3.3%of the original number of spectral bands(601),which greatly reduced the dimension of the spectral bands.Secondly,the three leaf vegetation indices RVI,DVI and NDVI at the three growth stages were constructed by random band combination,of which the best band combination of RVI was(636,678),the correlation coefficient was 0.8292,the optimal band combination of DVI was(676,515),the correlation coefficient was 0.9413,and the optimal band combination of NDVI was(681,639)and the correlation coefficient was 0.8444.In the"trilateral"parameter Dr andλr showed the largest positive correlation,the correlation coefficients were 0.856 and 0.826,respectively,and the Dy showed the largest negative correlation,and the correlation coefficient was-0.781.(4)The optimal spectral characteristic band,vegetation index and trilateral parameters were extracted as input vectors for the relative chlorophyll content inversion model of cotton canopy leaves of SVM,random forest and BP neural network,and the inversion model of relative chlorophyll content of cotton canopy leaves based on SVM,random forest and BP neural network was established.The results show that the relative content of chlorophyll in cotton canopy leaves at the bell stage based on random forest is the best modeling effect,and the determinant coefficient of the training set is R~2,the rms error is RMSE is 1.1857,the coefficient of determination of the verification set is R~2 is 0.8873,and the RMSE of the root mean square error is 1.5629.It can provide certain theoretical and technical support for quickly and accurately obtaining the chlorophyll content of cotton and monitoring the health status of cotton.