首页 > , Vol. , Issue () : 1993-2002
|沈宇||中国科学院遥感与数字地球研究所 再生资源实验室 北京firstname.lastname@example.org|
|李强子||中国科学院遥感与数字地球研究所 再生资源实验室 北京email@example.com|
|杜鑫||中国科学院遥感与数字地球研究所 再生资源实验室 北京|
|王红岩||中国科学院遥感与数字地球研究所 再生资源实验室 北京|
|张源||中国科学院遥感与数字地球研究所 再生资源实验室 北京|
玉米和大豆是两种主要的粮食作物，及时准确的监测两者的种植面积对于产量预测和市场价格的制定具有重要的意义。利用遥感技术探究在生长季中后期能有效区分玉米和大豆的指示性特征集，为在不同实验区进行推广应用和提前玉米和大豆种植面积信息发布的时间提供技术支撑。文章以玉米和大豆为研究对象，以黑龙江和安徽省两个典型种植区为实验区，以GF-1号影像为数据源，计算多种植被指数特征和两种纹理特征，同时利用特征优选方法评价特征间的相对重要性，并结合随机森林分类算法分析特征个数对精度的影响，得到不同试验区区分两者的最佳特征子集。随后根据不同实验区最佳特征子集的共同点和差异，遴选出对玉米和大豆中后期区分的遥感指示性识别特征集，并设计实验方案验证其有效性和稳定性。实验表明（1）在玉米和大豆生长中后期存在具有高效辨识两者的遥感特征集，能有效和稳定地增强两者的遥感识别能力；（2）在不同实验区，基于特征优选方法可以选择出区分玉米和大豆的最佳分类特征子集，得到两者最优的识别效果，比仅仅使用原始波段特征的分类精度提升了近10个百分点，总体分类精度能够平均达到97%，kappa系数0.96，玉米和大豆的单类分类精度平均超过95%；（3）在不同的种植区，利用玉米和大豆的指示性特征集可以得到几乎与优选出的最佳特征子集同样的分类精度和制图效果，且具有稳定性和有效性，较最佳特征集更具推广使用意义。指示性特征集包含6种：植被指数中的比值植被指数（RVI），差值植被指数（DVI），转换型植被指数（TVI），改进型叶绿素吸收比率指数（MCARI）和灰度共生矩阵（GLCM）纹理特征中的二阶矩（the Second Moment）和熵（Entropy）。
Corn and soybean are two main food crops. Timely and accurate monitoring of the planting area of both crops is of great significance for production forecasting and market price setting. The objectives of this article are to using remote sensing technology to explore the indicative feature that can effectively identify corn and soybean in the middle and later growth season, and provide technical support for the promotion and application of different experimental areas and the earlier release of corn and soybean planting acreage information. In this paper, two typical planting areas of corn and soybean were selected, the GF-1 satellite images acquiring in the middle and later growth stages were used as data sources to calculate a variety of vegetation index features and two textural features. At the same time, the features optimization method was used to evaluate the relative importance of features and select positive features for identifying corn and soybean, and next the random forest classification algorithm was utilized to analyze the relationship between the number of features and the classification accuracy, and then the best feature sets of different experimental areas could be obtained. Finally, according to the common points and differences of the optimal feature collections in different experimental areas, the remote sensing indicative feature sets for the differentiation of corn and soybean in the middle and later stages were found, and the next experimental scheme was designed to verify its validity and stability. Experiments show that (1) In the late growth stage of corn and soybean, there are features to identify the two crops highly efficient, which can effectively and stably enhance the remote sensing recognition ability of both; (2) In different experimental areas, based on the feature optimization method, the optimal feature sets for classification and the best mapping performance of corn and soybean could be both obtained, which is nearly 10 percentage points higher than the classification accuracy of using only the original band features. The overall classification accuracy can reach an average of 97%, the kappa coefficient is 0.96, and the single class classification accuracy of corn and soybean is more than 95% on average; (3) In different planting areas, using the indicative feature set of corn and soybean can obtain the same classification accuracy and mapping performance as the best optimal feature collection, and it was stable and effective, and it is more suitable for promotion and application. Indicative features include six types: ratio vegetation index (RVI) in vegetation index, difference vegetation index (DVI), conversion vegetation index (TVI), improved chlorophyll absorption ratio index (MCARI) and the second moment in Gray-level co-occurrence matrix (GLCM) texture features and entropy.