基于GF-1号影像的南方水稻种植信息提取

针对南方复杂地区水稻遥感信息提取研究中机器自动学习分类研究较少、分类精度不高的问题,以福建省三明市建宁县溪口镇为研究区,基于GF-1号卫星影像,采用面向对象的随机森林遥感分类算法对研究区内水稻田信息进行提取。首先通过优化面向对象分割参数和随机森林分类模型参数,提取并调用了影像中的多种特征;再对光谱特征、植被指数特征、纹理特征、几何特征进行特征空间优选;最后通过设置4种特征优选试验进行对比,得到最优分类模型。实验结果显示,基于特征空间优选的面向对象随机森林分类算法的水稻提取精度高达90%,分类总体精度可达87%,Kappa系数为0.85;与其他试验结果相比,漏分和误分现象较少,实现了南方地区水稻信息高精度自动识别。该方法计算特征少、实现简便,对于国产高分卫星影像在南方复杂地区作物自动提取中的应用具有参考性。

Information Extraction of Southern Rice Planting Based on GF-1 Image

In order to resolve the problem of insufficient study in machine automatic learning and low accuracy of remote sensing classification for the extraction of the rice in complex southern regions,taking Xikou Town,Jianning County,Sanming City,Fujian Province as the research area,we used an object-oriented random forest remote sensing classification method to extract rice information based on GF-1 satellite images in the research area.By adjusting object-oriented segmentation parameters and random forest classification model parameters,we used the characteristics of remote sensing images to optimize the characteristic space of spectral feature,vegetation index feature,texture feature and geometric feature,and set up four features of multi-selection test for comparison.Then,we obtained the optimal model of classification.The experimental results show that the extraction accuracy of rice based on the object-oriented random forest classification method reaches 90%,the classification overall accuracy reaches 87%,the Kappa is 0.85,and there are fewer missing classification and misclassification phenomena compared to other test results,which achieves high-precision automatic identification of rice information in southern region.This approach has few calculation features and is easy to implement,which can provide reference value for the application of high resolution satellite images in automatic crop extraction in complex southern regions.