Crop classification by random forest using TerraSAR-X data
山谷, 祐貴 ,
薗部, 礼 谷, 宏 ,
王, 秀峰 ,
小林, 伸行望月, 貫一郎
11 , 2017-03-30 , 北海道大学大学院農学研究院
This paper presents crop classification using satellite data to establish a mapping method in place of the existing ground survey. We calculated four variables of sigma naught, and polarimetric parameters from TerraSAR-X HH-VV dual-polarization data, and assessed the accuracy of classification performed by machine learning algorithm “Random Forest”. The result showed about 90% of accuracy when we used five dates’ imagery and four variables, respectively. And accuracy assessment was done under the condition when the number of variables or scenes was reduced. The accuracy became worse when the number of variables was reduced, but it can be maintained when the number of dates was reduced, thus these results confirm that crop classification with the lower cost will become possible.