Non-destructive Detection Of Hyperspectral Imaging In Detecting Potato External Defects

In this paper, the potato as the research object, on which the different types of surface defects were researched. Some external defects of potato caused serious influence on the quality. The traditional classification method is low efficiency, big labor intensity, poor objectivity and difficult to identify shortcomings.Using hyperspectral imaging technology of visible light to near infrared (400～1700cm), combined with image processing method, established the recognition algorithm of potato external defects, so as to realize the comprehensive evaluation of potato external quality and provide the theory basis for the development of online, real-time and fast nondestructive testing system during next step. The main research contents and results are as follows:(1) In order to realize the accurate and fast classification of potato in the process of actual processing, various potato external defects were detected in spectral region of400～1000nm using hyperspectral image technology, and an online non-destructive testing method was established by principal component analysis method of the characteristic wavelengths and image subtraction algorithm. Six defective potato types (mechanical damage, hole, scab, bruise, sprout, shagreen, normal) and one qualified potato type were used as the research objects in this study, and their hyperspectral images were obtained, respectively. Then the reflectance spectrums of the interested areas of potato in these hyperspectral images were extracted and analysed.(2) Principal component analysis was used for spectral data dimension reduction, selecting five feature wavelengths (478,670,723,819and973nm) according to the local extremums of weight coefficient curve of the second principal component image of all the potato types.(3) After that, principal component analysis was conducted again based on the five selected characteristic wavelengths, then elected the principal component images where the differences of grey value between the potato defect area and the surrounding area were most obvious. Potato external defects were identified through image processing methods, such as threshold segmentation, corrosion, expansion and connectivity analysis. The recognition rate of all the seven potato types using principal component analysis method of the characteristic wavelengths achieves82.50%.(4) Furthermore, In order to eliminate the impact of grey value of potato images in the background, and improve the contrast between defect area and the surrounding area, image subtraction algorithm was put forward in this study. Using the subtraction method based on the principal component image and the reference image can effectively eliminate the influence of background region and the reflect light of potato surface. Potato external defects were identified through image processing methods, such as threshold segmentation, corrosion, expansion and connectivity analysis. The recognition rate of all the seven potato types using image subtraction algorithm combined with principal component analysis method of the characteristic wavelengths is up to96.43%. This experimental results show that this image processing method based on hyperspectral image could identify the potato external defects effectively, so this research can realize the accurate and fast classification of potato in the process of actual processing and have a great application value in the future. (5) In order to explore the feasibility of on-line nondestructive testing on various potato external defects, a suitable detection method was developed based on hyperspectral image technology in nir spectral region of900～1700nm. Five defective potato types and one qualified potato type were used as the research object in this study, and their hyperspectral images were obtained, respectively. Then the reflectance spectrums were extracted from the interested areas in these hyperspectral images;(6) Principal component analysis was used for spectral data dimension reduction, and selected seven feature wavelengths (990,1026,1109,1226,1285,1464and1226nm);(7) Then principal component analysis was conducted again based on the selected the characteristic wavelengths, and elected the second principal component image that used in image recognition of potato external defects, with the recognition rating achieving71.25%.(8) In order to improve the recognition rate, band ratio algorithm combined with principal component analysis method of the characteristic wavelengths was put forward in this study, and the recognition rate was up to97.08%. The experimental results showed that the method based on near infrared hyperspectral image could identify the potato external defects effectively.