基于颜色特征的油菜缺素症图像诊断

摘要

To build an intelligent image diagnose technique,an intelligent diagnose of nutrients deficiency in rapeseed( Brassica napus L. )were proposed according to color variation characteristics of leaves from nitrogen, phosphorus and potassium deficiency. Using the Kawasaki soilless formula,rapeseed seedlings were cultivated and leaf images were collected. An image library was created with 4 types( normal,N deficiency,P deficiency and K deficiency). Template image with 4 gradients was set up using the GrabCut algorithms on significant color feature images. Others were subdivided into training and testing images. Matching indexes were calculated between with the training and template images using color histogram backprojection,which were utilized to train the Bayesian classifier for classification characteristics. Matching indexes of the testing images were computed into the classifier for obtaining the nutrients deficiency diagnostic of rapeseed. All algorithms were implemented with VC + + and Open CV. Results showed that the proposed method was able to accurately identify the common nutrients deficien-cy. Thus the above image diagnose could provide a constructive example for other nutrient deficiency diagnose.%为建立智能营养诊断技术，根据缺素导致的苗期油菜叶片颜色变化特征，提出一种甘蓝型油菜缺素计算机智能图像诊断技术。使用山崎配方无土培育了一批油菜并采集苗期油菜叶片图像，以建立正常、缺氮、缺磷和缺钾4类油菜图像库。使用GrabCut算法提取前景并挑选颜色特征显著的图像建立4个梯度的模板图像集合，其余图像被划分为训练图像和测试图像。训练图像对模板图像集合的颜色直方图反向投影得到匹配指数集合，用于训练贝叶斯分类器得到分类特征参数。计算测试图像的匹配指数并输入分类器得到缺素诊断结果。全部算法使用VC++和OpenCV实现。实验表明，该方法可以准确地判别常见的缺素状况，为基于图像处理技术的缺素诊断技术提供了一个有益的范例。