Abstract:
Abstract: Nondestructive detection of fruit ripeness is crucial for improving fruit's shelf life and industry production. Hyperspectral imaging is an emerging technique that integrates conventional imaging and spectroscopy to acquire both spatial and spectral information from a sample. It takes the advantages of the conventional RGB, near-infrared spectroscopy and multi-spectral imaging. In this work, hyperspectral imaging technology intended to determine a classifier that could be used for nondestructive classification for the ripeness of Shajin apricot. There were 480 Shajin apricot samples to be investigated, which were from an apricot planting garden in Xiaobai Village, Taigu County, and the samples were classified into 4 classes: unripe, mid-ripe, ripe and over-ripe according to the days after harvesting. Hyperspectral imaging technology with the band range of 400-1000 nm was used to evaluate nondestructively the ripeness of the Shajin apricot. The 480 RGB images were acquired for the apricot samples with 4 different ripeness classes (120 for each class). After acquiring hyperspectral images of Shajin apricot, the spectral data were extracted from the region of interests (ROIs). Using Kennard-Stone algorithm, all kinds of samples were randomly divided into training set (360) and test set (120) according to the proportion of 3:1. In this work, the soluble solid content (SSC) was chosen as an evaluation index of maturity for Shajin apricot. First of all, one-way analysis of variance (ANOVA) was used to evaluate the SSC of 480 samples of intact Shajin apricots at different ripeness stages. The results indicated that SSC presented significant differences among the different ripeness classes and had a increasing tendency along with the development of ripeness, which demonstrated that there was a high correlation between maturity and SSC with the correlation coefficient of 0.9386. Subsequently, based on the calculation of partial least squares regression (PLSR), 9 wavelengths at 434, 528, 559, 595, 652, 678, 692, 728 and 954 nm were selected as the optimal sensitive wavelengths (SWs), 6 statistical textural parameters of hyperspectral images including mean, contrast, correlation, energy, homogeneity and entropy were extracted by gray level co-occurrence matrix (GLCM) as the textural feature variables, and 6 statistical color indicators of hyperspectral images including mean values and standard deviations of R, G and B component image were extracted by RGB model as the color feature variables for the purposes of ripeness classification. Moreover, the ability of hyperspectral imaging technique to classify Shajin apricot based on ripeness stage was tested using the extreme learning machine (ELM) models. The ELM ripeness classification models were built based on the extracted SWs, texture, color, combination of SWs and texture, combination of SWs and color, combination of texture and color, combination of SWs, texture and color features, respectively. The results showed the correct discrimination rate was the highest for the prediction samples based on SWs and color features, and it reached 93.33%. The research reveals that the hyperspectral imaging technique together with suitable analysis model is a promising tool for rapid estimation of quality attribute and ripeness classification for Shajin apricot, which can provide a theoretical reference and basis for designing classification system of fruits in further work.