Feature Extraction And Classification Ofapple Mealiness Based On Hyperspectral Scattering Images

China is a big fruit production country. Because of lagging detection technique, fruit product industries are lack of competitiveness in the international market. The basic way to due with this problem is studying advanced fruit nondestructive testing technology. Hypersp- ectral scattering images technique which combines spectroscopy and image technique is suit to test fruit internal qualities and safety. Because of the large number of wavelengths and including spatial information, hyperspectral scattering image have large amount of data which will lead to process and analysis data difficultly. Here, the research of effective methods of selecting wavelengths and extracting features will promote the fruit nondestructive testing. Mealiness is an important indicator which reflects the internal qualities of apple fruit.This paper mainly studied the problem of extracting features of hyperspectral scattering images and classification of mealy apple samples by classification models. It provided the technical support for apple fruit real-time on-line detection and classification. The main contents can be completed as follows:1. This paper was aimed at evaluating and developing locally linear embedding (LLE) algorithm to extract spectral features from the hyperspectral scattering images for mealiness classification. LLE was developed to extract features from the hyperspectral scattering image data. Partial least squares discriminant analysis (PLSDA) was applied to develop classification models based on LLE and mean reflectance (MEAN) algorithms. The results show that compared with traditional MEAN algorithm, LLE algorithm provided an effective means to extract hyperspectral scattering features for mealiness classification.2. Uninformative variable elimination coupled with locally linear embedding (UVE-LLE) was used to assess apple mealiness. After the UVE, the number of effective wavelengths decreased to 23.5% of full spectrals of hyperspectral images. Then LLE was utilized to reduce the image data of effective wavelengths. The results showed that UVE-LLE have good classification accuracy. Compared with LLE, UVE-LLE could obtain the same classification accuracy but using small part of wavelengths of LLE model. Hence, UVE-LLE provides an effective algorithm for online classification and fast data saving of hyperspectral images data.3. The classification of apple mealiness is a pattern recognition problem in essential. This paper applied sparse representation classification (SRC) to assess apple mealiness. Considered the unbalanced samples of each class, genetic algorithm (GA) was utilized to deal with them. The results show that SRC coupled with GA could be applied as a new method for apple mealiness classification.