基于长时序遥感影像的珠三角水稻种植范围提取

Extraction of Paddy Rice Planting Range in the Pearl River Delta Based on Long Time Series Remote Sensing Images

以珠三角为例,采用阈值法和随机森林法对珠三角地区水稻进行提取,并比较提取效果。结果表明,阈值法提取的水稻数量的Kappa系数、总体精度、F1score、查全率等均明显低于随机森林法,但阈值法的查准率非常高。对随机森林输入特征进行组合发现,当输入的特征组合为光谱特征+植被指数+水体指数时,得到的分类精度指标结果最好,Kappa系数约为0.99。研究结果可为今后大范围的开展水稻提取、区分水稻与旱地作物、评估稻谷生产产量、水稻生产与管理提供参考。

Taking the Pearl River Delta for example,we used threshold method and random forest method to extract paddy rice in the Pearl River Delta region,and compared the extraction effects.The results show that the total accuracy,Kappa coefficient,F1 score and recall of paddy rice quantity extracted by threshold method are significantly lower than that extracted by random forest method.However,the precision of paddy rice extracted by threshold method is high.By combining the input features of random forest,it is found that when the input featurecombination is spectral features,NDVI and MNDWI features,the classification precision indexes are the best,and Kappa coefficient is about 0.99.The results can provide some references for rice extraction,distinguishing paddy rice from dry land crops,paddy rice yield evaluation,rice production and management in the future.