Nondestructive Detection And Quality Evaluation Of Chinese Chestnut Fruits Using Near Infrared Hyper Spectral Image Technique

Castanea mollissima is one of traditional export agricultural products in China which has high nutritional and economic values.It is known as woody grain,suitable for fresh food,stir-fry,vegetables,can also be used for processing Castanea mollissima powder,instant powder,as well as the production of Castanea mollissima paste,Castanea mollissima soup,and especialty,such as cakes.It is originated in China,generally can be divided into two categories that are the northern glutinous Castanea mollissima which rich in sugar and protein,and southern Castanea mollissima rich in starch.But,it often damage by chestnut as grade a pest in the Castanea mollissima production,the larvae feeding in chestnut leaf,and fecula full of decay within the tao but often difficult to detect in appearance.In addition,the depression of chestnut fruit for dense type seeds,can't stand the heat in the process of storage,afraid of cold dry,wet,it is difficult to get fresh-keeping effectively,and easy mixed fruit has lost commodity value and edible value in shipping period.Traditionally,Castanea mollissima detection using chemical analysis methods,more and more in the laboratory,the method to evaluate the broken after one by one in the fruit,time-consuming,low detection efficiency,and often limited test sample size,it is difficult to guarantee the sampling representative.To achieve fast nondestructive testing.Therefore,It is the current problems to develop a rapid,efficient,non-destructive fruit detection technology,in order to meet the needs of the fruit grading a large-scale quality analysis and to improve standardization level of fruit by post-processing.With the rise of hyper spectral image technology that is the unity of multi-bands,high resolution and mapping characteristics of images and spectroscopy analysis can fuse the various advantages.At the same time,as a kind of rapid,efficient and nondestructive detection method,has been rapid development in detection of agricultural products,medicine,chemical industry.But due to high dimension characteristics of hyper spectral itself,different methods of dimension,and spectral data processing method of difference causes the diversity of model.Therefore,it requires us to optimizate fruit fast nondestructive testing method based on the analysis of the hyper spectral image technology.This experiment begins with Castanea mollissima as materials try to establish Castanea mollissima fruit internal quality and external criterion of plant diseases and insect pests by combining near infrared hyper spectral image technology with different chemical metrology nondestructive testing methods.This experiment,through continuous optimization model,in the final choice PLS(partial least squares)to establish a quantitative model of index of Castanea mollissima,can be used to predict Castanea mollissima fruit inclusion content;identifying the distinction between the different varieties of Castanea mollissima and plant diseases and insect pests by using Discriminant analysis method and combining with the method to establish the qualitative model for differentiating chestnut fruit variety identification and plant diseases and insect pests,to obtain the test conclusion is as follows.Choosing 'talil'Castanea mollissima as sample,use DA method establishting qualitative discrimination model to determine the mildew,insect pests and normal state.When using SNV+log10+SG method,model recognition rate reached 98.6%,with a strong applicability.First,using DA method to set up three chestnuts qualitative model with'tailil','yi meng duan zhi','shu cheng xiao li ' Castanea mollissima.And found the model of using MSC+ log10+SG smoothing pretreatment method with the best results of the discrimination rate of 96.7%.Illustrate the applicability of the model.Secondly,through PLS modeling method,forecast analysis two important indicators of total sugar and starch on three kinds of Chinese chestnut for the samples that 'taili 1','yi meng duan zhi','shu cheng xiao li' Castanea mollissima.Compare different spectral preprocessings of Castanea mollissima index prediction results and found that MSC+second derivative+SG smooth forecast results are good method,model to predict the results of correlation coefficient reaches the lowest is 0.9313,highest 0.9587,the root mean square error minimum 0.0624,the highest for 0.225,model predicted results is very considerable.Third,choose 'tailil'Castanea mollissima as sample,use DA method establishting qualitative discrimination model to determine the mildew,insect pests and normal state.When using SNV+log10+SG method,model recognition rate reached 98.6%,with a strong applicability.