基于GF-1数据和非监督分类的冬小麦种植信息提取模型

针对当前冬小麦种植信息提取方法普遍存在严重依赖地面样本数据和人为主观干扰过多的现象,而非监督分类算法自身又具有独特的特点,研究了基于非监督分类的冬小麦提取方法。在实际应用中,非监督分类的初始分类数目难以准确确定,这会导致分类精度降低或分类结果需要进行二次人工合并。通过时间序列曲线和差值增强技术解决了初始分类数目难以准确确定的问题,提出了一种以归一化植被指数为冬小麦信息识别指标,基于高分一号数据和非监督分类的冬小麦种植信息提取模型。以河北省辛集市为研究区,应用该模型提取了2014和2015年辛集市冬小麦种植信息,并应用混淆矩阵方法进行精度验证和与监督分类方法对比分析。结果表明:①该模型冬小麦的制图精度为94.23%～96.64%,用户精度为92.31%～95.45%,Kappa系数0.89,整体精度达到94.33%以上;②在无需地面样本数据支持的条件下,该模型可以达到近似监督分类的提取精度。可见提出的冬小麦种植信息提取模型精度较高,可以满足区域内冬小麦种植信息地面遥感监测的需求,是一种行之有效的冬小麦种植信息提取新方法。

Extraction Model of Winter Wheat Planting Information Based on GF-1 Data and Unsupervised Classification

Aiming at the fact that the extraction methods of winter wheat planting information generally rely heavily on the ground sample data and human subjective interference, this paper studied the extraction methods based on unsupervised classification algorithm which has its own unique characteristics. In practical application, it is difficult to accurately determine the initial classification number of unsupervised classification, which will lead to the reduction of classification accuracy or the need for secondary manual merging of classification results. This paper solved the problem through time series curve and difference enhancement technology, and proposed a new extraction model of winter wheat planting information based on GF-1 data and unsupervised classification, which takes normalized difference vegetation index (NDVI) as identification index. In this paper, the winter wheat planting information in Xinji City of Hebei Province in 2014 and 2015 was extracted by the model, and the accuracy was verified and compared with the supervised classification method by the confusion matrix method. The results show that: (1) the producer accuracy of the model is 94.23%-96.64%, the user accuracy is 92.31%-95.45%, the kappa coefficient is 0.89, and the overall accuracy is over 94.33%; (2) without ground sample data, the model can achieve the extraction accuracy of approximate supervised classification. It can be seen that the model has relatively high precision and can meet the needs of remote sensing monitoring of winter wheat planting information in the region. So, the model is an effective new method of winter wheat planting information extraction.