紫茎泽兰是一种原产于墨西哥的世界性恶草，从上世纪40年代传入我国，现已侵入我国西南大部分省份，每年以10-30km速度继续向北、向东扩散。紫茎泽兰有毒，且有很强的繁殖能力，对农、林、牧业都有很大的危害。准确、高效地获得紫茎泽兰的分布信息对紫茎泽兰的扩散研究以及治理预防等都有重要的意义。本研究以西昌市作为研究区域，获取2011全年HJ-1星的CCD影像，构建年度NDVI时间序列数据集，通过对数据集的处理分析，结合地面样点，依据紫茎泽兰的NDVI年度变化特征，采用监督分类和决策树的方法提取紫茎泽兰的分布信息。本研究的工作和成果主要有以下几个方面：（1）基于HJ-1星CCD数据的时间序列NDVI数据集构建。本研究首先对下载的原始数据进行几何校正、辐射校正、裁剪镶嵌等一系列预处理工作，计算NDVI。对获取的NDVI数据采用最大值合成方法，分别获得时间分辨率为16d和32d的时间序列数据集。通过对比分析，发现16d的数据可以更好的反映植被在不同生育期的变化特征。在16d分辨率数据集的基础上，选用HANTS软件对数据进行滤波重建，进一步削弱云、大气、地形等影响，获得年度NDVI数据集。（2）分析不同地表类型年度NDVI变化曲线，提取紫茎泽兰识别依据。根据野外调研及文献资料收集方法，获得典型植被样点，提取不同植被的年度NDVI曲线。通过对比分析发现：农田NDVI时间谱线呈双峰变化模式，与当地农作物主要是两季播种相符；林地呈单峰变化模式，冬季NDVI仍处于比较高的值，主要是由于林地以常绿阔叶林、针叶林等树种为主；草地呈单峰模式，秋天NDVI很小；紫茎泽兰谱线介于林地、草地之间，与另外两类最明显的差别是在3、4月份NDVI有比较明显的降低，这主要是由于3、4月份是紫茎泽兰花期，紫茎泽兰花是白色，会导致NDVI值变小，是识别紫茎泽兰的重要依据。（3）采用监督分类和决策树分类方法提取紫茎泽兰分布信息。监督分类的样本主要是依据野外调研、文献资料及高分影像。决策树主要考虑紫茎泽兰生长海拔、年度NDVI均值、峰值变化、花期特点来确定决策树规则。两种方法都有比较好的分类精度，监督分类精度高于决策树，但两种方法对林下紫茎泽兰的识别有一定的局限性。

Crofton weed is one of the world's malignant weeds native to Mexico, which has been introduced into China since 1940s. Crofton weed has invaded most of the provinces in southwestern China, and continued to spread to the north and east. Crofton weed is poisonous and has strong reproduction ability. It has great harm to agriculture, forestry, and animal husbandry. Therefore, getting the spatial distribution of the Crofton weed accurately and efficiently has important significance to the diffusion research and governance with prevention of the Crofton weed.In this paper, Xichang City was taken as a study area, of which CCD images of full-year 2011 were obtained from the HJ-1 data, then the annual NDVI time-series data set were built. Further, using the supervised classification and the decision tree method, the spatial distribution of the Crofton weed was finally obtained by analyzing of the data set, and ground samples, based on the annual NDVI characteristics of the Crofton weed.The main research works and conclusions of this paper are as follows:(1) The NDVI time-series data was established based on the CCD record of the HJ-1satellite. To calculate the NDVI data, a series of pretreatment to the original data were firstly carried out, including geometric correction, radiation correction, clipping mosaic, etc. Using the maximum synthesis method, time-series data sets with temporal resolution of 16 days and 32 days respectively were obtained from the NDVI data. Through comparative analysis, it is found that time-series data sets with temporal resolution of 16 days could better reflect the variation characteristics of vegetation in different life cycles than the one with temporal resolution of 32 days. Based on the data set with temporal resolution of 16 days, using the HANTS filter reconstruction technology, effects such as the cloud, the atmosphere, and the terrain were further weakened, and then the annual NDVI data sets were obtained.(2) Recognition basis of the Crofton weed was obtained by analyzing different annual NDVI variation curves of different vegetations. According to the collected samples from field survey and literature data, different annual NDVI time-series of different vegetations were extracted. Through comparative analysis, it is found that NDVI seasonal trajectories of farmland presented two growing seasons per year, in accordance with the local crops planting situation of two seasons. The NDVI seasonal trajectories of woodland showed one growing season per year, but the NDVI still stayed at a relatively high value in winter, which is mainly due to the trees of evergreen broad-leaved forest and coniferous forest are the majority trees. The NDVI seasonal trajectories of grassland also showed one growing season per year, and value of NDVI begun to decrease in autumn, and reached a very small value in winter. The variability of NDVI seasonal trajectories of the Crofton weed is between that of the woodland and grassland. The most obvious feature of the Crofton weed NDVI curve is the significant reduction of NDVI value in March and April, this is mainly because that March and April is the flowering period of the Crofton weed, of which flowers are white, leading to the decline of NDVI value, and this is an important basis for the identification of the Crofton weed.(3) The supervised classification and the decision tree method were used for recognition of the Crofton weed distribution. The samples for the supervised classification were on the basis of field campaign, previous studies and high resolution image. The decision tree was created by determining the rules of the decision tree, considering the growth altitude, the average of annual NDVI, the change of growth pattern and the flowering characteristics. The two methods above could both recognize the Crofton weed with good classification accuracy. The supervised classification has an advantage over the decision tree. However, they both have some limitations for recognition of undergrowth Crofton weed.