成像光谱数据特征选择及小麦品种识别实验研究

本文选取中国科学院栾城农业实验站的实验田块（分类结果如图7）进行了精度

503小麦 9204小麦 4185小麦 早熟小麦 菜地 绿化地 砖瓦厂 乡村小路 居民点 验证。根据农业站田块工程设计图和实地丈量结果，得出每种小麦的面积（单位平方米，见表2），再从分类结果图中计算出每种小麦分对和分错的面积，从而计算出实际分类精度，见表2，结果表明，小麦品种分类精度，除4185外均达到90％以上，总的小麦分类精度超过90％。

图7 实验点成像光谱数据分类结果 Fig.7 the classification of the test area

表2 四种小麦的分类精度

Table 2 the classification accuracy of four wheat types

503小麦 9204小麦 4185小麦 早熟 合计 503小麦 12996 117 0 0 13111 9204小麦 459 10953 621 0 12033 4185小麦 0 468 6849 315 7632 早熟小麦 162 72 378 4536 5148 其他地物 383 390 152 149 1074 合计 14000 12000 8000 5000 39000 分类精度% 92.8 91.3 85.6 90.7 90.6

4 结 论

高光谱数据提供了丰富而又冗余的信息，本文以国产成像光谱仪数据为例，对高光谱数据的特点进行了分析，引用遗传算法结合小麦生物物理特性筛选出了参与分类的波段。利用Fuzzy-Artmap分类器进行了地物分类，识别了不同的小麦品种。我们认为利用GA算法结合小麦生物特性进行的特征选择从理论上是可靠的，从实践上是可行的，是一种针对作物品种分类的有效特征选择方法；对不同小麦品种识别的精度超过90％，说明利用高光谱数据结合合理的特征及分类器的选择，能够实现不同小麦品种的识别。

参考文献 (References)：

[1] Lakshmana Rao K M. Remote Sensing for

Land-Use Planning [J] International Journal of Remote Sensing, 1995, 16(1): 53-60.

[2] Tucker, C. J., Townshend, J. R. G., and Goff, T. E..

African land-cover classification using satellite data［J］. Science, 1985, 227:369-375. [3] Defries, R. S. and Townshend, J.R.G., 1994,

NDVI-derived land cover classification at global scales［J］. International Journal of Remote Sensing, 15:3567-3586.

[4] Staenz K., J.-C. Deguise, J. Chen, H. McNairn, T.

Szeredi, M. McGovern.,1998,The Use of Hyperspectral Data for Precision Farming［C］; ISPRS Commission VII Symposium, Budapest, Hungary, September 1-4, pp. 38-42.

[5] Staenz K., 1996,Classification of a Hyperspectral

Agricultural Data Set Using Band Moments［J］. Canadian Journal of Remote Sensing, Vol. 22, No 3, pp. 248-257

[6] Deguise J.-C., M. McGovern, H. McNairn, K.

Staenz.,1998,Spatial High Resolution Crop Measurements with Airborne Hyperspectral Remote

9

Sensing［C］. 4th International Conference on Precision Agriculture, Saint Paul, Minnesota, 19-22 July, pp. 1603-1608.

[7] Tong Qingxi, Zheng Lanfen, Wang Jinnian etc,

1997, Imaging spectrometry remote sensing study for wet land vegetation［J］. Journal of Remote Sensing in China, 1(1):50-57.［童庆禧，郑兰芬，王晋年等，1997，湿地植被高光谱遥感。遥感学报，1（1）：50-57。］

[8] Gong Peng, Pu Ruiliang, Yu Bin,1998,

Hyperspectral data identify and analysis for different temporal conifer breeds［J］. Journal of Remote Sensing in China, 2(3):211-217.［宫鹏，浦瑞良，郁彬，1998，不同时相针叶树种的高光谱数据识别和分析。遥感学报，2(3):211-217.］ [9] Foroutan, I. and Sklansky, J., 1987. Feature

selection for automatic classification of non-gaussian data［J］. IEEE Transactions on Systems, Man and Cybernetics, 17: 187-198. [10] Fukunaga, K., 1990. Introduction to Statistical

Pattern Recognition［M］. Academic Press, New York.

[11] Richards, J.A., 1993. Remote Sensing Digital

Image Analysis: an introduction［M］. [16]

Springer-Verlag Berlin Heidelberg, Second Edition. [12] Bruce, L. M., Koger C. L., and Li J., 2002.

Dimensionality Reduction of Hyperspectral Data Using Discrete Wavelet Transform Feature Extraction［J］. IEEE Transaction on Geoscience and Remote Sensing, vol.40(10): pp.2331-2338 [13] Lee, C., and Landgrebe, D. A., 1993, Feature

extraction based on decision boundaries［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence, 15, 388- 400.

[14] Yang Jiangang, 2001, Artificial neural network

applied tutorial[M]. Zhejiang University Press.[ 杨建刚，人工神经网络实用教程。浙江大学出版社，2001。]

[15] Zhou Zhihua, cheng Zhaoqian, Cheng Shifu, 1999,

Self-adaptive resonance theory review［J］. Computer science, 26(4): pp54 –56.[ 周志华，陈兆乾，陈世福，自适应谐振理论综述。计算机科学，1999，26(4):：pp54 -56。]

A Experimental Study on Imaging Spectrometer Data Feature Selection and Wheat Type

Identification

Wang Chang-yao1, Liu Zheng-jun2, Yan Chung-yan3

（1. Key laboratory of Remote Sensing Information Science, Institute of Remote Sensing application, Chinese Academy of Sciences, Beijing 100101, China

2. Institute of Photogrammetry and Remote Sensing, Chinese Academy of Survey and Mapping, Beijing 100039, China

3. China University of Geosciences, Beijing, 100083, China)

Abstract:

With the development of imaging spectrometer technology, the ground objects’ consecutive information it provided make it possible to identify different vegetation types, though some relevant research was carried out in the past few years, most are about forestry, yet few about crops. Further,

10

there exist strong correlation between bands of imaging spectrometer, so how to reduce as much as possible the redundant information and reserve useful information appear much more important. This paper first did feature selection based on genetic algorithm(GA) and wheat biophysical characteristics. In feature selection using GA, for the training samples, when combined bands reach 4, the JM distance of optimal combination reach much high level, when bands go on increasing, the average JM distance increases slowly until when bands reach 8, the distance increases no more, so the optimal bands combination can be obtained. In feature selection using wheat biophysical characteristics, we found that there appear strong correlative bands for wheat protein and dry gluten with spectra, so the sensitive bands can be obtained. Combining these two feature selection steps, the ultimate optimal bands combination was given. After feature selection, we use the selected bands and classifier Fuzzy-Artmap to classify the imaging spectrometer data. It showed that for 4 wheat types, they can be identified clearly, the average classification accuracy is above 90% . . Key words: imaging spectrometer, feature selection, type identification

第一作者简介：

王长耀，男，（1941.1-），1993年获奥地利Graz大学博士学位，中科院遥感所博士生导师，长期从事农业与生态环境遥感应用研究，获得过国家及部级多项科学技术奖励，出版和发表过多项专著和论文，其中SCI文章3篇。

11