Abstract
This study was conducted to recognize powdery mildew disease on the leaves of cucumber cultivated in greenhouses using digital image processing and pattern recognition techniques. The image analyses of the leaves were done using the image toolbox in MATLAB. Given an original image in RGB color space, two kinds of normalized images of a mean-normalized image and an index-normalized image were transformed from the original image. RGB(red, green, and blue) components and HSI(hue, saturation, and intensity) color features were separated from original image and two normalized images respectively. The performance of the disease recognition were analyzed in related with an infected area calculated from the separated image. In order to test the effect of thresholding method, two optimal thresholding methods (Otsu's method and ISODATA algorithm) were applied to the all images which have RGB components and HSI color features separated from the original image, the mean-normalized image, and the index-normalized image. Optimal thresholding method is the way of choosing the threshold value automatically. The ISODATA algorithm resulted in overall better recognition of powdery mildew disease than Otsu's method in the case of bin=2. New image was obtained to test the effect of array operation of segmented images by array multiplication. The array multiplication could eliminate unnecessary information such as vestiges of powdery mildew disease, discoloration by other diseases and light. Also, Median filter operation was effective for noise elimination and reduction of image distortion.
Figures & Tables
Fig. 1. Flow chart of disease detection algorithm
