Fast Prediction Of Moisture Content Of Potato Planting Leaves In Field By Near Infrared Spectroscopy

With the development of modern agricultural mechanization,the situation of large-scale concentrated planting has become more and more mainstream,but it has also led to the inability to timely and comprehensively cover the supervision of plant growth.Therefore,the requirements for modern agriculture tend to be more precise and efficient.In terms of crop irrigation,modern agriculture requires fast and accurate acquisition of the water content of plants and the water information of plants in different regions,which provides powerful help for regional precision irrigation.In terms of detailed management of crops in agriculture,it can also save manpower and water resources.As one of the main crops in the world,potato has the largest production scale in the world in my country,and it plays a powerful role in world food security.During the growth and development of potato plants,water plays a vital role.The cell activity of leaves is closely related to the water content of leaves,and cell activity affects the rate of photosynthesis.Therefore,the lack of water will affect the growth,yield and quality of potato.Therefore,finding a fast,simple and accurate method to detect leaf moisture content has become the focus of attention.The appearance of hyperspectral imaging technology provides a new measurement method for the detection of plant moisture.Near infrared spectroscopy provides the absorption information of the combined frequency and multiple frequency of the vibration of the hydrogen-containing groups in molecules,which can show the data characteristics of the hydrogen-containing groups in the sample.This technology has the advantages of fast,accurate,concise,non-destructive,non-contact,etc.,and has been widely used in the field of agricultural product testing in recent years.Therefore,near infrared spectroscopy can be used to predict the moisture content of agricultural products.In view of this,the thesis carried out research on the prediction of potato leaf moisture content based on hyperspectral imaging technology.Firstly,the spectral reflectance information of 110 fresh potato leaves in the band range of 900～2100nm was collected by hyperspectral imaging system.Due to the serious noise impact and large data redundancy in spectral data processing,in order to achieve the precision and efficiency required by modern agriculture,it is necessary to denoise the data and extract characteristic bands in the paper.In terms of denoising,four preprocessing methods:S.G smoothing,multivariate scattering correction(MSC),standard normal variable transformation(SNV),and mean centering are used.After processing the original spectral data,partial least squares are established.Regression(PLSR)model,BP neural network model and least squares support vector machine(LS-SVM)model,and then based on the prediction results,screen out the pre-processing method and modeling method with the best prediction effect.The results show that the recognition rates of the established models are all over 70%.Among them,the BP neural network model established by multivariate scattering correction preprocessing has the best predictive effect.The predictive set determination coefficient R~2 is 0.9791,and the root mean square error RMSE is0.3723.In terms of characteristic waveband extraction,three characteristic wavelength extraction methods are used:Regression Coefficients(RC),Principal Components Analysis(PCA),and first derivative method.After processing the original spectral data by three methods,three prediction models are established respectively.Multiplicative regression model,BP neural network model and least square support vector machine model,and then based on the prediction results,screen out the feature wavelength extraction method and modeling method with the best prediction effect.The results show that the recognition rates of the established models are all over 60%.Among them,the BP neural network model established by the characteristic wavelength extracted by the regression coefficient method has the best prediction effect.The prediction set determination coefficient R~2 is 0.9698,and the root mean square error RMSE is 0.3177.Finally,considering the agricultural application of data acquisition is outside the field,the experiment on the basis of the above the indoor test results is relatively complete,the data collecting site moved to outdoor,and based on the multiple scattering correction pretreatment method,Regression coefficient method(Regression Coefficients,RC)characteristic wavelength extraction method,BP neural network modeling method to establish a complete model for predicting water cut in outdoor green leaves.The results show that the prediction effect of the model is good,the determination coefficient R~2 of the prediction set is 0.6352,and the root mean square error RMSE is 1.2269,which can realize the basic outdoor prediction,and provide ideas and methods for the automation of modern agricultural machinery.