DEVELOPMENT OF CLASSIFICATION TECHNIQUE FOR POTATOES USING IMAGE PROCESSING

Loading...
Thumbnail Image
Date
2020-04-22
Journal Title
Journal ISSN
Volume Title
Publisher
MPUT, UDAIPUR
Abstract
Potato (Solanum Tuberosum L.) being the most important crop, farmers don’t get good supporting price due to lack of improper grading methods. To fill up this gap the present research has been carried out with the help of image processing (non destructive technique). The setup developed to capture images has a resolution of 0.22 mm/pixel. Cropped images have major and minor diameter resolutions of 0.47 and 0.53 mm/pixel respectively. With the help of image processing, adapted and improved feature extraction techniques (gray level co-occurrence matrix properties, histogram of oriented gradients, local binary patterns, gabor and radon) of five different types of potatoes i.e. crack, rotten, sprout, skin peel and good potato features were extracted and subsequently compared with various input parameters. In adapted method of HOG feature extraction, the feature vector lengths were found higher compared to the improved method for all potato classes. For crack and rotten potato images, HOG visualization time in improved method was higher compared to adapted method and for remaining potato image classes; adapted method had more HOG visualization time. The LBP feature vector lengths in improved method were higher compared to adapted method for five potato classes and the squared error’s for all potato classes were same. The gabor feature vector length was higher in adapted method compared to improved method for all potato classes. Improved method required less time to plot gabor magnitude and spatial kernels for all potato classes. For all potato classes, the radon feature vector sizes were same in both adapted and improved methods but differ in plotting times of radon transformation. In improved method, less time was taken for plotting radon transforms while contrast values were higher compared to adapted method for all five potato classes. In adapted method, correlation values and energy values were greater than adapted method for all potato classes. Also, in adapted method, homogeneity values were greater for all five potato classes compared to improved method. The developed classification data set was optimised using artificial neural network application toolbox in “Matlab” at mean squared error of 1.4899 after five iterations. The neural network algorithm had to run the classification data set to five times to achieve minimum squared error. An attempt was made to classify potatoes based on gray level co-occurrence properties. Cubic kernel SVM was found to have highest 113 accuracy with 99.5 per cent compared to quadratic kernel, linear kernel, medium Gaussian kernel, fine Gaussian kernel and coarse Gaussian kernel using one versus one classification method. Accuracies of cubic kernel SVM was found highest for one versus one and one versus all classification methods.
Description
Development Of Classification Technique For Potatoes Using Image Processing
Keywords
Citation
Dhulipalla R.B. And Verma R.C.
Collections