The Technologies For Nondestructive Detection Of Moldy Core In Apples Using Transmission Spectroscopy

The detection of moldy core in apples and its disease degree in apple moldy core,has important application value in guiding the orchard sterilization treatment,avoiding disease development,improving fruit quality,enhancing market share and so on.Domestic and foreign scholars have carried out extensive and in-depth research,and the beneficial exploration for detection of apple moldy core.However,there are still some key problems to be solved: although widely used in the detection of internal quality of apple,the reflectance spectrum detection method is not suitable to detect internal quality of apple,and the advancement and accuracy of the detection model are also to be improved.In addition,the factors that affect the spectral transmission of apple moldy core have not yet been determined,the detection of the disease degree of apple moldy core is uncertain,and type detection of apple moldy core has not yet been reported.For the above-mentioned problems,this paper aims to find out the detection factors that affect the spectral transmission of apple moldy core,improve the detection accuracy of existence or nonexistence of apple moldy core,construct discriminant model and method of the disease model and disease degree of apple moldy core.This research includes modeling method of support vector machine(SVM)and deep belief network(DBN),SVM of which parameters are optimized by genetic algorithm(GA),and the design and implementation of the nondestructive detection system of apple mold disease using transmission spectroscopy.The main contents and conclusions of the paper are as follows:(1)The influencing factors research of apple moldy core detected using transmission spectroscopy is carried out,and the main factors of influencing the detection of moldy core are determined,and the better parameters of collecting transmission spectrum are obtained.Because the early stage of apple moldy core occurs in core and its surrounding,the non-destructive detection method of apple moldy core with the transmission spectrum is used,through the establishment of detecting experimental platform,adopting broadband light source to irradiate apple to be measured,and receiving transmission spectrum.The factors that include the type,intensity,quantity,installation position,shading mode and spectral range of the light source are studied,and the better parameters for collecting thetransmission spectrum are obtained.(2)The dimension reduction methods for high-dimensional spectral data suitable for the detection of apple moldy core are studied and determined.For the problem that the data band are complex,overlapping and large in data collected by spectrograph,the correlation between the specific spectrum in the whole band scope and the disease degree of apple moldy core is established.Studying dimension reduction methods for high-dimensional spectral based on principal component analysis(PCA),successive projections algorithm(SPA)and wavelet analysis,the partial least squares discriminate analysis(PLS-DA)and SVM optimized by GA(GA-SVM)detection models of apple moldy core are established to carry out test,respectively.The results show that three feature extraction methods(PCA,SPA and wavelet analysis)can effectively reduce the data processing dimension and improve the modeling efficiency in the case of guaranteeing the accuracy of the prediction model,and the average discrimination time of the PLS-DA sample based on the SPA feature extraction method is about 10% time-consuming of PLS-DA model based on full spectrum data.In addition,the feature extraction of wavelet analysis reduces the data dimension and eliminates the partial noise signal at the same time,and the prediction accuracy of PLS-DA and GA-SVM models based the feature extraction of wavelet analysis,is 84% and 97.33%,respectively.(3)A discriminated method based on GA-SVM for the existence or nonexistence and the disease type of apple moldy core was proposed.For the characteristics and problems of nondestructive detection of moldy core in apples using transmission spectroscopy,discrimination model of apple moldy core with PLS-DA and BP neural network(BP-NN)of which the weights and threshold value are optimized by GA(GA-BP-NN)were constructed.Since the traditional linear modeling method cannot map the nonlinear relation and the accuracy of the BP neural network model is to be improved,the SVM models of discrimination model of apple moldy core were established.However,without theoretical guidance,the penalty coefficient and kernel function parameter not easy to choose in SVM model,so GA is used to optimize SVM model parameters because of parallelism and strong search ability of GA.Compared with PLS-DA and GA-BP-NN models,the results showed that the prediction accuracy of GA-SVM model for detection of moldy core in apples is97.33%,and the prediction accuracy of GA-SVM disease symptom types detection model of apple moldy core is 81.48%,which shows that the GA-SVM model has higher prediction accuracy and can achieve accurate prediction of the moldy core disease of Red Fuji apples.(4)A discriminative method of the disease degree of apple moldy core based on DBN was proposed.In order to realize the non-destructive testing of the disease degree of apple moldy core,the disease degrees of moldy core were divided into healthy,mild,moderate and severe in accordance with the proportion of decay area in intersecting surface.The PLS-DA,BP neural network and SVM model were studied and established for recognition test,and then on this basis,in order to improve the accuracy of the disease degree of apple moldy core,the discriminative method of the disease degree of apple moldy core based on DBN was studied based on the theory of deep learning.Through the theoretical analysis and experiment,the DBN network structure which is suitable for the detection of the disease degree of apple moldy core was established,and the DBN network training method was studied.The experimental results show that the accuracy of disease degree of DBN discriminative model is 88%,which is 5.33 percentage points higher than that of PLS-DA,BP-NN and SVM,and has better recognition effect.(5)The non-destructive detection system for apple moldy core using transmission spectrum was developed.The nondestructive detection system was realized based on USB2000+ spectrometer experimental platform,using Java development language and OmniDriver development kit,by calling the GA-SVM and DBN discriminative algorithm.After testing and verification,the detection accuracy rate of apple moldy core reached96.67%,and the detection accuracy rate of the disease degree reached 87.5%,the effectiveness of identification method and detection system have been verified.