Rapid Detection and Visualization of Mechanical Bruises on 'Nanfeng' Mandarin Using the Hyperspectral Imaging Combined with ICA_LSQ Method

摘要

The hyperspectral imaging (HSI) is used in this work to detect slight mechanical bruises (indentation) on "Nanfeng" mandarin (NM) surface. The independent component analysis (ICA) and the least square method (LSM) are used for data dimension reduction and determination of the coefficients corresponding to the images at characteristic wavelengths, respectively. The Lambertian model and the mean normalization method are used for spectral data preprocessing before applying the ICA, and then, six images centered at the wavelengths of 556 nm, 578 nm, 593 nm, 604 nm, 616 nm, and 683 nm are selected by analyzing the separating matrix of the ICA4. Subsequently, the LSM is used to calculate the coefficients of the ICA4 to enhance the visual contrast of a combination image between the indentation and sound region on the fruit surface. During detecting the indentation on a fruit, the stem-end and calyx of the fruit can be misclassified as indentation, so another six images centered at wavelengths of 556 nm, 604 nm, 647 nm, 670 nm, 716 nm, and 819 nm are selected based on the ICA2 to segment the stem-end and calyx before detecting an indentation region. The experimental results show that performance of the image segmentation algorithm is acceptable, the total success rate for indentation and sound fruits is 93.98%, and the total success rate for stem-end and calyx is 100%, indicating that the proposed multispectral algorithm is capable of detecting the indentation on fruit. The images at the characteristic wavelengths can also be used to establish an online multispectral imaging system for detecting the indentation regions on fruits.