自然条件下酿酒葡萄病害检测方法的研究

项目名称： 自然条件下酿酒葡萄病害检测方法的研究

项目编号： No.61461005

项目类型： 地区科学基金项目

立项/批准年度： 2015

项目学科： 无线电电子学、电信技术

项目作者： 冯全

作者单位： 甘肃农业大学

项目金额： 43万元

中文摘要： 病害的发生是制约酿酒葡萄产量和品质的重要原因。大部分病害发生时会在葡萄叶片上产生相应症状。利用视觉检测技术对葡萄叶片上病斑进行检测和识别，实现病害检测的自动化，对病害的早期、快速、准确检测和诊断有重要意义。 在西北，对自然状态下的酿酒葡萄叶面病害的检测是一项极富挑战性的任务，叶面尘土沾附、光照变化、叶片相互遮挡等因素对病害准确检测带来困难。本项目研究了尘土引起的图像退化对病害检测的影响和恢复模型的建立，研究颜色恢复方法以应对不同光照；根据形状、纹理、颜色等先验知识为叶片和多种典型葡萄病害建立鲁棒的检测模型和分割方法；为了适应实际中病害表现的多样性，研究灵活的分类器的在线学习和模型更新方法；研究序列图像中叶片的跟踪定位问题，利用常见噪声与病斑在时间空间相关的信息方面之差异去除噪声；还研究近红外主动成像条件下叶片以及病害检测，以及利用近红外与彩色图像的联合信息对彩色图像中叶片分割方法。

中文关键词： 尘土退化模型；颜色恢复；近红外光；在线学习；时空联合去噪

英文摘要： Diseases seriously affect the production and quality of wine grape. When most grape diseases occur, the corresponding symptoms are found on the leaves. Computer vision can be employed to detect and identify disease spots of the leaves, which is of great significance to rapid and accurate detection of diseases in early periods. In the Northwest China, it is a great challenge to detect diseases on a grape leaf under natural conditions - the dust adhering to a leaf, varying sunlight, mutual occlution of leaves, etc., which all prevent an accurate detection. The effects of dust on the disease detection are researched and the corresponding model of restoration is built. The methods of color restoration are also studied to deal with different illuminations. The robust methods of detecting and segmenting a leaf and some representive diseases of wine grapes are researched with respect to the prior knowledge of shape, texture and color. Since the deseases show different forms in pracitce,the smart classifiers for the diseases capable of learning online and updating are studied. The methods of leaf location and track in sequental images are studied, as well as the ways of denoising by utilizing the differences of spatio-temporal information between disease spots and typical noises. Near-Infrared (NIR) images are employed to detect the leaves and diseases, and NIR and color information are integrated to segment leaves better.

英文关键词： Degradation Model of Dust Image;Color Restoration;Near-Infrared;online learning;spatio-temporal denoise