Study Of Optimal Model For Nondestructive Detection Of Kiwifruit, Peach And Pear Quality Characteristic By Nir Spectroscopy

The production and consumption of fruit in China has been in the front rank of the worldfor a long time, nevertheless, there is a big disparity in quality and processing products offresh fruit between China and some developed countries, which is the main bottleneckimpeding the development of the fruit industry and also prevent our fruit industry fromentering into the world.The major reasons for this gap is that traditional detecting techniquescan not master the quality change of fruit in many processes,such as during its growth period,storage period, intermediate phase, processed phase and so on accurately, rapidly andcomprehensively. Thus, it is of great practical importance to detect the fruit quality usingnovel nondestructive measurement techniques.Under the base of research group, the analysis models of detecting soluble solids contentof mixture variety fruit, identifying fruit variety and recognition bruised kiwifruits based onnear infrared diffused spectroscopy technology were established with the experimentalmaterials of kiwifruit, peach and pear, the crucial technical methods used in process of themodeling analysis were also compared and evaluated systematically and comprehensively,and relatively optimal analytical models were finally determined. The main results andconclusions are shown as follows:(1) The move window partial least squares and least squares support vector machine(MWPLS-LSSVM) model for detecting soluble solids content (SSC) of mixture varieties ofkiwifruit combined with sample set partitioning based on joint X-Y distances (SPXY)algorithm partitioning sample sets and D1lg(1/R) method preprocessing original spectra had arelatively optimal prediction precision and feasibility. The correlation coeffiencient ofcalibration (Rc), root mean square error of calibration (RMSEC), correlation coeffiencient ofprediction (Rp), root mean square error of prediction (RMSEP) of the model were0.977,0.641,0.948and0.762respectively.(2) Sample sets were partitioned by Kennard-Stone method, original spectra were used,and the models established by Fisher, BP network, LSSVM and ELM coupled with successiveprojection algorithm (SPA) selecting characteristic wave numbers for identifying varieties ofkiwifruits (Huayou and Xixuan no.2) had relatively optimal discriminant performance with an accuracy rate of100%for all prediction kiwifruit samples.(3) Sample sets were partitioned by Kennard-Stone method, original spectra werepreprocessed by multiplicative scatter correction (MSC) method and the model established bySPA-LSSVM for recognition of bruised kiwifruit had relatively optimal discriminantperformance with an accuracy rate of100%,85.7%and76.8%for identifying collidedsamples, pressed fruits and intact ones, and the discriminant accuracy rate for total sampleswas88.1%.(4) The uninformative variable elimination and LSSVM (UVE-LSSVM) model fordetecting SSC of mixture varieties of peaches combined with SPXY and original spectra had arelatively optimal prediction precision and feasibility. Rc, RMSEC, Rpand RMSEP of themodel were0.998,0.220,0.985and0.428respectively.(5) Sample sets were partitioned by Kennard-Stone method, original spectra were used,and the models established by Fisher, BP network, LSSVM and ELM coupled with SPA foridentifying varieties of peaches (Beijing no.8, Laishanmi, and Shahong) had relativelyoptimal discriminant performance with an accuracy rate of100%for all prediction peachsamples.(6) The UVE-LSSVM model for detecting SSC of mixture varieties of pears combinedwith concentration gradient method and MSC preprocessing original spectra had a relativelyoptimal prediction precision and feasibility. Rc, RMSEC, Rpand RMSEP of the model were0.986,0.163,0.974and0.262respectively.(7) Sample sets were partitioned by Kennard-Stone method, original spectra were used,and the models established by Fisher, BP network, LSSVM and ELM coupled with SPA foridentifying varieties of peaches (Dangshansu and Xuehua) had relatively optimal discriminantperformance with an accuracy rate of97.5%for all prediction pear samples.