Study On Detection Methods Of Defect Detection For Potato Based On Transmittance And Reflectance Hyperspectral Imaging Technology

As one of the important crops in the world, potatoes have good industrial, edible and medicinal value. The quality of potato, especially its defects will directly affect the economic value. Potato defect detection research helps to improve the commercial value. In practical testing process, the extent of the defect, the orientation of the defect, the critical parameters of detection system and data processing methods will all be of great impact on test results, so study on the multi-parameter of detection system, determine the various factors of materials and the influence of different data processing methods on the test results, then build the internal and external nondestructive defects detection model of potato based on the transmission and reflection hyperspectral imaging technology, to achieve rapid and accurate detection, so as to adapt to the demand of the real-time online detection, and it has great scientific significance and good application prospects.This paper take the potato produced in Shanxi province, China as the research object, successfully build the transmittance hyperspectral imaging acquisition system. With the integrated use of the transmittance hyperspectral imaging technology, reflectance hyperspectral imaging technology, image analysis technology, spectral analysis technology and data analysis technology, study on qualitative identification method for inside black heart especially minor black heart potato and external injury of the arbitrary placement of potato. Firstly, study on the related parameters of the transmittance and reflectance hyperspectral imageing cquisition platform and then determine the parameters; Sencondly, compare the detection accuracy of potato black heart based on transmittance and reflectance hyperspectral imageing techniques and determine the transmittance hyperspectral imaging technology enables detection of mild black heart of potato; Finally, do detections for external injury potato randomly placed based on the transmittance and reflectance hyperspectral imageing technology, compare the detection accuracy of external injury potato about the reflectance image, reflectance spectrum and transmittance spectrum, build the optimal detection model for external injury potato randomly placed.The results are as follows:1) Build a hyperspectral imaging acquisition platform of transmission and reflection, study the related parameters of them.Determine the light source for the reflectance hyperspectral imaging system is an annular light source, and the best image acquisition speed is2mm/s. Study on the edge intensity, gradients, image information entropy, standard deviation, and the degree of distortion and spectral correlation of the reflected test hyperspectral images, the results show that, images acquired with the ring light is better; the best image acquisition speed is2mm/s with theoretical analysis combining experimental studies.Determine the optimal light intensity for the transmittance hyperspectral imaging system is3halogen lamps of50W, and the best image acquisition speed is2mm/s.3halogen lamps of20W,50W,100W have been respectively used, the intensity of the transmitted under three hyperspectral images of transmittance spectral is analyzed to determine the optimal3halogen lamps of50W; the best image acquisition speed is2mm/s with theoretical analysis combining experimental studies.2) Determine the optimal pretreatment method is transform trends, best modeling approach is partial least squares for reflectance hyperspectral imaging data, accuracy rate of discriminant model is89.47%, in which the recognition accuracy of black heart samples is84.38%, the accuracy of the normal samples is96.00%.3) Determine the optimal pretreatment method is standard normal variable correction, the optimal modeling method is partial least squares for transmittance hyperspectral imaging data, accuracy rate of discriminant model is96.49%, in which the recognition accuracy of black heart samples is97.30%, the accuracy of the normal samples is95.00%.4) Compare the detection accuracy of black heart potato with transmission and reflection hyperspectral imaging techniques, and determine the best hyperspectral imaging technology is transmission hyperspectral imaging technique.5) Study on the model optimization methods of potato blackheart detection based on the transmission hyperspectral imaging technology, determining CARS-SPA is the best. Use MC-UVE, SFLA, SPA, CARS, CARS-SPA and other variable selection methods for the transmittance spectrum variable selection, and ultimately determine the CARS-SPA is the best, build the PLS-DA model by the5variables selected by the CARS-SPA, accuracy of model is96.49%overall, in which the identification accurate of the black heart sample is97.30%, the identification accuracy of normal samples is95.00%.6) Study on external damage detection methods of randomly placed potato based on image dimension of reflection hyperspectral imaging technology. Apply IC to extract the characteristics of reflectance hyperspectral image, do the second IC analysis on the reflectance image, and build the identification model based on the reflectance image. The results show that different orientations have great influence on identification accuracy, the recognition accuracy is highest when damage facing positive to the camera, in which the mechanical damage is90.91%, the bruised is93.10%, and the normal sample is94.00%.7) Study on external damage detection methods of randomly placed potato based on spectral dimension of reflection hyperspectral imaging technology. Apply IC to extract the characteristics of reflectance hyperspectral image, do the second IC analysis on the reflectance image, and build the identification model based on the reflectance spectral, The results show that recognition accuracy for bruised is higher, the recognition accuracy is96.55%when the damage is facing positive to the camera, back to the camera is94.83%, side to the camera is91.38%. But the identification accuracy of mechanical damage is low, in which the highest is only78.18%.8) Study on external damage detection methods of randomly placed potato based on spectral dimension of transmission hyperspectral imaging technology. Apply IC to extract the characteristics of reflectance hyperspectral image, do the second IC analysis on the transmittance spectrum, and build the identification model based on the transmittance spectrum, The results show that recognition accuracy for bruised is higher, in the three directions the recognition accuracy is all100%; the recognition accuracy is100%when the mechanical damage is facing positive and back to the camera, and side to the camera is98.18%.9) Study on the model optimization methods of potato damage detection based on the transmission hyperspectral imaging technology. Apply sub-window arrangement analysis algorithm (SPA) for further choice of transmission spectra and use3spectral variables to establish identification PLS-DA model of randomly placed potato damage, the accuracy of the model is97.39%overall. The results show that this study could provide the technical support for on-line detection of randomly placed potato damage with transmittance hypersensitive imaging technology, and the recognition accuracy is higher than that with reflectance hyperspectral imaging technology.