Research On Internal Quality Detection Of Fruits Using Spectrum And Imaging Technology

Detection and classification of fruit is closely related to enhance economic benefits offruit grower and businesses, to meet consumer demand for high-quality fruit. The past20years, the near-infrared spectroscopy and hyperspectral imaging technology become animportant fruit quality detection technology which are rapid, nondestructive and can detect avariety of indicators. NIR and hyperspectral techniques for fruit quality detection have somekey factors affecting accuracy, such as wavelength selection, feature extraction, andappropriate model. The main contents in this paper are as follows:1. The number of NIR wavelength is too large, and prediction model is difficult toguarantee generalization. For this problem, the paper proposed a new method IGA-PLSP toselect NIR wavelength, for citrus soluble solids content (SSC) forecasts. Firstly, the immunegenetic algorithm (IGA) is used to select sub-interval, on this basis, use partial least squaresprojection algorithm select wavelength furtherly. The results show that: IGA-PLSP can selecteffective wavelength to improve the prediction accuracy of the model, and greatly reduce thenumber of wavelengths.2. This paper proposes a new method of extracting hyperspectral scattering imagefeatures, which combine generalized gaussian distribution (GGD) and the average reflectanceintensity. The new method considers the amplitude of reflectance intensity, at the same time,use GGD fit the spatial distribution of reflectance intensity, which can extract image fullinformation. PLS was used to predict firmness and SSC of apple. Compared with averagereflection method (Mean) and modified lorentzian distribution (MLD), the features extractedby GGD-Mean can improve prediction accuracy of firmness and SSC.3. The evaluation of fruit taste parameters is fuzzy, the research proposes a classifierbased on fuzzy relation (FR) for apple mealiness discrimination. The method uses theprinciple of FR convert two instruments indicators to membership of apple mealiness. Then,PLS is used to establish apple mealiness classification model. The results show that: newmethod can improve the classification accuracy.