| Potato staple food is Chinese main strategy in production of potato.Potato will become the fourth staple food in China.Quality testing is an important part of potato seed breeding,production,processing,storage and sales.Potato quality testing technology research has played a significant role in ensuring the development of potato industry.Currently,potato quality testing is based on non-destructive testing technology both at home and abroad,and machine vision methods and near-infrared spectroscopy methods are most common method.A series of research progress has been made on potato external defect detection technology based on machine vision and spectroscopy,but there are few studies on potato scab detection methods.Therefore,in order to improve the recognition rate of potato scab disease and achieve rapid and non-destructive detection of scab disease,the main research contents of this paper include:1.Study a method based on the Gaussian Laplacian feature for the location of the potato scab.By using the sensitivity of the Gaussian Laplacian to the edge,all the edge features of the potato body were extracted.The dense regions in the feature matrix were extracted by the dilate method and K-means clustering methods to obtain the position information of the potato surface.Positioning the stains can avoid the interference of the sore stains such as mechanical damage and scatter spots on the identification of the surface of the potato directly.2.A dynamic potato deblurring and enhancement method with moderate degree of ambiguity was studied to reduce the image blur caused by frame rate,exposure time and so on.Comparing and analyzing the degree of blurring of dynamic potato photos at different motion rates.The dynamic image was sharpened by Laplacian method,and the image to be detected image was deblurred to improve the recognition rate of dynamic potato scab.3.Study the dynamic image enhancement method based on linear gray-scale variation and cubic spline interpolation.The sharpened image is enhanced to obtain more effective pixel points,which solves the problem of effective pixel reduction when shooting dynamic images.The linear gray-scale transformation method enlarges the gray-scale difference between the potato body and the stain,enlarges the number of effective pixels,and highlights the sharpness of the image of the spot.The spline interpolation method is used to modify the gradient-incremented partial pixel value introduced by the gray-scale transformation,and the introduction of artificial noise is reduced while enhancing the image.4.Study methods for identifying potato scab based on concentric contours,roundness and pixel area values were studied.The scab is similar in shape to the ring,and the result of a scab spot extraction has multiple concentric contours,which can be used to screen exclude most non-scarred areas.Exist the small part of the spot has multiple contour,which the roundness low with high area,and the roundness of the large but area is very low.Therefore,the regression equation of roundness and area can be used for detect characteristic parameters.Segmentation is performed to identify the lesions of the scab.5.Study the scab identification model based on regression segmentation curve and BP neural network(back propagation).Three characteristic parameters of roundness,area and centroid coordinate corresponding of potato scab were extracted.Concentric contour detection and segmentation curve were used to establish the scab disease recognition model based on regression segmentation curve.The correct rate of flea disease potato identification was 91.37%.After the characteristic parameters were normalized to different scales,to establish BP neural network scab disease identification model,the correct rate of scab disease detection was 96.25%. |
