Comparative analysis of Indian wheat seed classification

In this study 2-layer ANN (artificial neural network) a linear classifier and k-NN (k-nearest neighbor) a non-linear classifier were applied for identification and classification of images of four Indian wheat seed species into four classes of wheat seeds on the basis of their varieties. 120 images (40 images of four classes, 10 images of each class) from three different places were taken under same illumination condition. These images were cropped to 320×240 resolution and converted into gray scale images. We had extracted 131 texture features of wheat species using various textural algorithms which contain LBP(local binary pattern),LSP(local similarity pattern),LSN(local similarity numbers), GLCM(gray level co-occurrence matrix),GLRM(gray level run length matrix) matrices of gray image. The feature group which gave highest percentage of accuracy in classification was determined. The determined feature group showed maximum average accuracy of 100% for inter-class classification and 66.68% for intra class classification when it was classified linearly i.e. using ANN. On the other hand it gave 85% of average accuracy for inter-class classification and 39% for intra class classification with non-linear classification method, using k-nearest neighbor (k-NN). Thus, results shows the linear classifiers are outperformed to non-linear one as features are linear in nature.