中文摘要：

为准确快速定量诊断黄瓜的病害，科学选择病害管控措施，基于Android技术和图像处理方法设计了可用于自然背景的黄瓜叶部病害定量诊断系统，并进行了试验。对黄瓜叶部彩色图像，首先进行图像预处理和背景剪除，再识别病斑区域，最终计算病斑区域占其所在叶片区域的百分比及根据国家相关标准与其对应的病害等级，计算结果以数值形式显示在诊断结果界面，同时用红色标识出病害区域。系统既适用于白色打印纸等简单背景，也适用于较为复杂的自然背景；所识别的病害叶片图像既可以从摄像头实时获取，也可以从手机存储载入。以50幅黄瓜霜霉病病害叶片为对象对系统进行测试，试验结果表明，系统可以较准确地对黄瓜霜霉病病斑区域进行识别（病斑区域识别综合误分率为6．56％），并按照国家标准给出病害等级（综合错误分级率为3％）；简单人工背景下系统识别时间为1s，自然背景下系统识别时间约为11s。

英文摘要：

Accurate and rapid disease severity quantifying is critical for scientific selection of disease control measures. Smartphone-based systems may facilitate this procedure. Based on Android and digital image processing, a smartphone-based system for cucumber leaf disease severity quantifying was designed and implemented. Leaf images can be obtained by using the smartphone back camera in field, and also can be loaded from local storage of the smartphone. Severity quantifying was done to the image in several steps. Firstly, image pre-proeessing and non-interested background removal were directly done to the leaf color image. Secondly, the diseased region was discriminated from the leaf region. Finally, disease severity was calculated by the ratio of disease area to leaf area as percentage, and disease grade was also calculated from the disease severity following a national standard. Numerical severity quantifying results were displayed in the interface, and the identified diseased region of the leaf image was marked in red and displayed in the interface as a synthesis image simultaneously. Two background removal algorithm were implemented in the system. One was used for simple background removal, namely super-G, which was used for background removal when the leaf region within a simple artificial background, such as a white A4 sheet. The other one was grabcut, which was a user-interactive background removal method chosen for complex natural background removal. Where the user could roughly point out background and foreground, and then the application would do the rest. For testing performance of the system, totally 50 images of downy mildew infected cucumber leaves were used. Images were acquired from greenhouses in north of Beijing. Results showed that the system could accurately quantify the downy mildew disease severity in acceptable time. Average percentage of false quantifying was 6.56%. Average running time for disease severity quantifying was 1 s for disease images with simple artificial backgrounds and 11 s （