不同特征信息对TM尺度冬小麦面积测量精度影响研究

充分挖掘遥感数据信息，改善作物识别环境，一直是农作物遥感监测的重要工作。以往研究表明最佳波段组合、纹理信息和植被指数信息可以在一定程度上提高分类精度，但这些手段是否一定可以提高作物识别的精度，不同分类器对不同特征信息组合的响应是否一致等都是值得探讨的问题，也是目前研究甚少的问题。为此，该文将平均值(Mean)、方差(Variance)、均一性(Homogeneity)、反差(Contrast)、相异性(Dissimilarity)、熵(Entropy)、角二阶矩(Angular Second Moment)、灰度相关(Correlation)7种纹理信息以及比值植被指数(RVI)、土壤调整植被指数(SAVI)、重归一化植被指数(RDVI)、植被液态水含量指数(NDWI)、有效叶面积植被指数(SLAVI)5种植被指数信息分别加入到TM多光谱数据中，同时还进行了最佳波段选择，利用最小距离、最大似然和支持向量机3种方法进行分类提取小麦，研究了不同特征信息对小麦测量精度的影响。结果表明：该试验区内最佳波段5、4、3组合，纹理信息和植被指数信息的加入，对小麦面积测量精度的提高没有贡献；同一个特征信息组合对不同的分类器影响不同。在实际小麦面积测量的操作中，作业员不应该盲目的加入特征信息。选用何种信息不仅仅和研究区本身的性质有关，还和使用的分类器有关。

Effects of various feature information on the accuracy of winter wheat planting area measurement

It is always an important piece of work to fully mine remote sensing image information and improve identification environment in agricultural crops remote sensing monitoring. Ancient study indicates optimal bands combination, texture and vegetation indices can advance classification accuracy in a certain extent. However, whether they can contribute to improve the crop identification accuracy out of question and whether they have identical response to different classifiers. These above problems, which are very important and valuable in agricultural crops area monitoring, are currently less researched. Hence, in this paper, seven types of common texture and five vegetation indices were respectively added into TM multispectral bands to classify using three different methods, which are Minimum Distance, Maximum Likelihood and Support Vector Machine, and analyze the effect on winter wheat identification accuracy by comparing the classification results. The contexture include Mean, Variance, Homogeneity, Contrast, Dissimilarity, Entropy, Variance, Angular Second Moment and Correlation, and the vegetation indices are RVI, SAVI, RDVI, NDWI and SLAVI. Results show that the optimal bands combination(band 5th, 4th and 3th), contexture and vegetation indices do not contribute to advance the wheat area measurement accuracy in this area. Same feature information combinations have diverse response to different classifiers. So, the interpreters should not blindly use feature information in wheat planting area measurement. How to choose the appropriate feature information is related to not only study area characteristics but also classifier.