中文摘要：

为提高黄瓜霜霉病叶部病害机器自动识别的准确性和鲁棒性,提出了一种基于视觉显著性图的黄瓜叶部霜霉病识别方法。首先将图像从RGB色彩空间变换到HSV色彩空间中进行色彩修正,再变换回RGB空间利用R、G、B分量的线性组合生成视觉显著性图,最后通过对生成的视觉显著性图进行阈值分割以识别病害区域。利用从北京市北部郊区日光温室采集到的50幅具有典型霜霉病特征的黄瓜叶片原始图像进行实验,结果表明,该方法能较为准确地从叶部彩色图像中识别出霜霉病病斑区域,平均误分率为6.98%,优于K-means法（11.38%）和OTSU法（15.98%）;平均运行时间0.661 4 s,少于K-means法的1.424 9 s;运行时间的均方根误差为0.051 5 s,鲁棒性优于K-means硬聚类算法。

英文摘要：

In order to increase the efficiency and robustness of automatic recognition of cucumber downy mildew disease,a disease recognition method was proposed in the fashion of visual saliency. Firstly,image sample of RGB color space was transformed into HSV color space,and a color correction method was performed on the sample image. Then the color-corrected image was transformed from HSV color space back to RGB color space,and a linear combination of the R,G,B components was carefully chosen to generate visual saliency map of disease area on the leaf image. Finally,based on the visual saliency map,the disease area was extracted from the leaf area of original image. 50 samples for testing were acquired from warm houses in northern Beijing from September to October,2015. Samples were taken by consumer grade digital cameras and mobile-phones with camera module. In order to focus on the problem of disease recognition,original leaf images＇ background were removed manually and uniformly fitted into 512 pixel by 512 pixel squares before experiments. Result of testing shows that this method can effectively extract disease area from color image with relatively high accuracy,the average of misclassification rate is 6. 98%,better than K-means（ 11. 38%） and OTSU（ 15. 98%）; the average running time is 0. 661 4 s,faster than K-means（ 1. 424 9 s）; the RMSE of running time is 0. 051 5 s,robuster is better than K-means. Result also shows that CC（ Color correction） method makes better results than original proposed disease recognition method proposed,mis-classification rate was decreased from 8. 63%（ Saliency method） to 6. 98%（ CC ＋ Saliency method）.