Research On Sensitive Band Extraction Method Of Rapeseed Canopy Spectrum Based On Multifractal Theory

As an important organ for photosynthesis,the plant canopy can reflect the growth of large area crops.Nowadays,spectrum technology makes it possible to non-destructively monitor crop and provides real-time diagnostic advice.Therefore,the canopy spectrum is one of the ideal targets that studies the rapeseed's growth monitoring and nutritional diagnosis.But the original canopy spectral acquisition easily is affected by the natural environment and human factors,the traditional methods are difficult to effectively extract spectral characteristic which contains rich growth information for the plant.Multifractal method is an important tool to extract feature and widely be used in various fields.In this paper,we proposed a novel sensitive bands extraction method for rapeseed canopy spectrum based on the multifractal theory.And we established SPAD prediction model and identification model based on the novel characteristics.The aim is to provide theoretical basis for real-time monitoring and nondestructive diagnosis of crops by combining physical methods and mathematical methods.1.The paper has realized the extracting method based on multiscale multifractal detrended fluctuation analysis(MMA)of rapeseed's canopy spectrum sensitive bands and has further established SPAD prediction model and identification model based on Hurst index of sensitive bands.The experiment shows that the sensitive bands are roughly stable at 534nm-574 nm by comparing to three parameter groups of the MMA method.Three evaluation indicators of root mean square error(RMSE),correlation coefficient(R)and relative error(RE)employed in the SPAD prediction model and identification model demonstrate that our Hurst exponent has the best performance.2.On the basis of traditional extraction method for red edge parameters,we have proposed a new method to acquire the rapeseed's spectral red edge parameters based on local detrended fluctuation analysis(L-DFA).In practice,we have calculated the local Hurst exponent of spectrum by using the L-DFA firstly.And then,six traditional red edge parameter extraction methods are employed to act on the estimated local Hurst exponent.Thus,four red edge parameters,i.e.red edge position(REPH),red edge amplitude,and left and right red edge area are obtained.In our experiments,the above four local Hurst exponent-based red edge parameters as argument,a prediction model is developed to forecast the SPAD values(Soil and plant analyzer development).The results show that the local Hurst exponent-based red edge feature can produce better model effect than the original spectrum-based one for the all phenological periods.Significance test for the four kinds of plant patterns shows that there is huge difference of the proposed four-dimensional red edge features between the transplant and direct plant in the whole process of rapeseed's growth.