论文部分内容阅读
Food safety has become a great public concern, and food poisoning or food-borne illnesses can lead to disturbance to the society.Residual pesticides in vegetables and fruits are one of the typical food safety issues around the world.Consequently, fast and nondestructive method is required for detecting the residual pesticides for food safety.As an emerging technology, hyperspectral imaging has been successfully employed in a large number of food safety inspection and control.In general, a hyperspectral scene is over multiple object;and each pixel of the observed spectra usually comprises multiple spectral signatures especially in the biological tissues.So, hyperspectral unmixing with the ability to identify individual material from a hyperspectral scene is critical for hyperspectral data exploitation.Spectral unmixing is one of the most prominent research in mining spectral information, and a large number of methods belong to spectral unmixing are unsupervised with no prior knowledge.Using spectral unmixing methods, pure components and their distributions can be derived.In this paper, hyperspectral imaging coupled with spectral unmixing methods was applied for detecting the specie of pesticide, qualitative analysis of the residual on the surface of leaves, and also semi-quantitative determine the relative level of pesticide on leaves.Two species of common pesticides, dichlorvos and omethoate were chosen, and each of them was diluted with deionized water to prepare a series of standard solutions (1∶1, 1∶5, 1∶10, 1∶20).Tea leaves were selected as the contaminated object, and regions where the diluted pesticides smeared were determined in order to assess the final results.Then each of these solutions was evenly smeared in the regions at the decided sequence.The Vis-NIR hyperspectral imaging system consisted of a spectrograph, a 12-bit CCD camera, and two 150W tungsten halogen lamps were used to acquire the hyperspectral images of the contaminated leaves.We chose the region of interest (ROI) at the experimental object for the further study.Several spectral unmixing methods were applied to analysis the data cube from hyperspectral images, consequently the spectra of pure components and the distribution of each component were acquired.Through comparing the extracted spectra with the reference spectra of the pure pesticide at the characteristic bands, the species of pesticides could be determined.The distinct and clear boundary could be present at the contaminated area, which were matched with the regions decided before.The abundance plots allowed one to appreciate the proportion of endmembers contained in the different regions of the image, so they could reflect the relative solution density on the surface, and the recognition accuracy reached 83.3%.This combination will open a new view to recognize the contaminated area of the plant or other foods nondestructively, also to visualize the degree of the contaminations.This study indicates that hyperspectral imaging combined with spectral unmixing analysis can be considered as an alternative technique for conventional methods in realizing online inspection, leading to the elimination of the occurrence of the problems about residual pesticides on the surface of leaf.