Crop Salt-alkali Stress Detection Method Based On Near Infrared Spectroscopy

Saline-alkali stress is one of the key abiotic stresses affecting crop yield and quality,At present,the detection of salt and alkali stress of crops is mainly through the traditional chemical analysis method.In order to avoid the complex operation and the lack of damage and time-consuming,a rapid non-destructive detection technology of salt and alkali stress degree of crops is studied to solve the problem of salt and alkali stress degree detection of field crops,It is of great significance and practical value to ensure the healthy growth of crops.In this paper,taking kidney bean as the research object,the batch spectral data of hydroponic kidney bean in different salt stress periods are collected,the data preprocessing is realized by analyzing multi-dimensional spectral characteristics,the characteristic wave number of spectral curve is extracted and optimized,and the intelligent detection model of salt stress degree is established,which is based on the research of crop salt and alkali stress detection method by near-infrared spectroscopy.The main research contents are as follows:(1)Spectral data acquisition.Taking the "HYD" of kidney bean varieties obtained by Heilongjiang Academy of Agricultural Sciences as the research object,Through the method of hydroponics and sodium bicarbonate solution of 100 mmol / L to simulate salt alkali stress,salt alkali stress was carried out when a compound leaf of kidney bean seedling was fully unfolded.According to the physiological indicators of kidney beans: photosynthetic pigment content,gas exchange parameters and chlorophyll fluorescence parameters were measured every 24 hours to determine the level of saline-alkali stress,and the period of health(no saline-alkali 0h),24 h,48h,72 h,96h,120 h and 144 h saline-alkali stress(canopy has lost its activity at 168h).By collecting canopy layers of kidney beans with different levels of saline-alkali stress,the corresponding near-infrared spectral data were obtained,and the spectrum data of kidney beans obtained in each period were numbered to accurately correspond to their saline-alkali stress levels.A total of524 sets of sample data were recorded.(2)The spectral data is preprocessed.For the canopy spectral data of kidney beans with different saline-alkali stress levels,a combination of principal component analysis(PCA)and Mahalanobis distance(MD)was selected to remove abnormal samples,and the originalreduced-dimensional data when the principal component was 3 was used to replace the original spectral data,then the Markov distance method was used to remove abnormal samples,and the results showed that there were no abnormal kidney bean samples.Then use the SPXY method to divide the spectral sample set,and apply convolution smoothing(SG),multiple scattering correction(MSC),standard normal variable transformation(SNV),standardization(STD),detrending(DT),and mean centralization(MC)methods to preprocess the original spectrum,which shows that the detrending preprocessing method works best.Its root mean square error(RMSE),correlation coefficient(r),determination coefficient(R2)and standard error(SE)averaged 0.685,0.945,0.895 and 0.675,respectively.(3)The characteristic wave number of the spectral curve is extracted and optimized.After DT preprocessing,the competitive adaptive reweighted sampling algorithm(CARS),successive projections algorithm(SPA),and CARS-SPA algorithm were used to extract the characteristic wavenumbers of the spectral curve of kidney bean canopy were 128,11 and 12 respectively.The total number of spectral bands was reduced by 91.25%,99.25%,and 99.18%,respectively,which greatly reduced the dimension of the spectral data.Then,based on a variety of modeling,the optimal characteristic wave number algorithm is determined.The characteristic wave numbers are extracted based on the three algorithms of CARS,SPA,and CARS-SPA.Through comprehensive analysis of the evaluation parameters of the separately established PLSR,PCR,and MLR models,the CARS algorithm is selected.As the optimal algorithm for extracting the characteristic wave number of the spectral curve.(4)Build a neural network model for intelligent detection of saline-alkali stress and evaluate the results.BP and RBF were used to comprehensively analyze the evaluation parameters of the six types of saline-alkali stress intelligent detection models established by the original spectrum,DT pre-treatment spectrum,and CARS extracted characteristic spectrum.All of them are better neural network models established by CARS extracted characteristic spectrum.The constructed DT-CARS-RBF model of 128-282-7 type three-layer neural network has better performance,and its simulation time,detection accuracy and mean square error are 30 seconds,98%,and0.00999354,respectively.The detection model constructed by DT-CARS-RBF method shows good fastness and accuracy.This study uses near infrared spectroscopy processing technology combined with a variety of intelligent information processing methods to construct a kidney bean salt-alkali stress detection model(DT-CARS-RBF),which can provide technical support and reference for field crops to automatically detect salt and alkali stress degree quickly and accurately.