基于光谱和形状特征的水稻扫描叶片氮素营养诊断

使用扫描仪获取水稻叶片图像,综合运用数字图像处理技术、参数优选和分类方法,研究了不同氮素水平水稻叶片的光谱和形状特征,并进行了氮营养的诊断与识别。研究利用面向对象的分类方法提取叶尖部位的黄化面积比例,指数回归分析结果显示此参数与叶片氮含量具有很高的相关性(R2=0.863)。提取整叶和叶尖的颜色参数并分别与叶片氮含量进行指数回归分析,发现叶尖部位的颜色特征能更好地反映叶片的氮素营养状况。采用CfsSubsetEval和Scatter search相结合方法对特征进行约简与优化,根据选择结果结合支持向量机方法进行模式识别。精度检验结果显示该方法对缺氮和正常叶片的正确识别率较高,随氮素水平的升高,正确识别率降低,对过量水平的正确识别率较低,叶面积在缺氮和正常模式下能对识别起到很好的辅助作用。

Diagnosis of Rice Nitrogen Nutrition Based on Spectral and Shape Characteristics of Scanning Leaves

The leaves of rice were captured by scanner.Integrated method combining digital image processing,parameter optimization and classification was used to explore leaves spectral and shape characteristics which were adopted to diagnose and recognize rice nitrogen nutrition.Proportion of etiolated area in the tip of leaf was extracted by method of object-oriented classification.The results of exponential regression analysis showed high correlation between tip etiolated area proportion and leaf nitrogen concentration(R2=0.863).The color indices of tip as well as whole leaves were extracted and exponential regression analysis with leaf nitrogen concentration was made,which illustrated the better performance of representation of rice nitrogen nutrition with tip information.Optimal selection of subset by means of CfsSubsetEval and Scatter search combined with support vector machine were used for pattern recognition.The result of accuracy assessment indicated that nitrogen deficiency and healthy leaves could be easily recognized and the accuracy descended with the improvement of nitrogen treatment.The accuracy of excessive nitrogen nutrition status was low.The leaf area could be a favorable assistant for recognition under deficient and healthy status.