红外, 2019, 40 (3): 24, 网络出版: 2019-04-11  

基于决策树和神经网络的农作物分类研究——以廊坊市为例

Crop Classification Research Based on Decision Tree and Neural Network——Take Langfang City as an Example
李龙 1,2,*李旭青 1,2吴伶 3杨秀峰 1,2孙鹏飞 1,2
作者单位
1 北华航天工业学院,河北 廊坊 065000
2 河北省航天遥感信息处理与应用协同创新中心,河北 廊坊 065000
3 中国地质大学(北京),北京 100083
摘要
以河北省廊坊市永清县整个县域为研究区,以GF1-WFV 16 m分辨率影像为数据源,选取覆盖作物完整生长期多个时相的影像数据,构建作物归一化植被指数(Normalized Difference Vegetation Index,NDVI)时间序列。通过对研究区NDVI曲线的分析,发现利用该数据构建的NDVI时间序列可描述研究区作物的生长特性,体现当地不同作物的物候差异,能有效地区分出当地的种植模式。选取NDVI曲线上最大值、最小值、峰值的出现时间、峰值数量和阈值等特征参数构建决策树。根据研究区的物候历和对当地种植结构的调查,利用最佳时相的影像,针对某一种或特定几种作物进行分类提取。分别采用决策树分类、神经网络分类等方法进行精度验证,综合比较得出最佳的作物分类方法。研究结果表明,在永清县这一县域研究区,利用GF1-WFV 16 m分辨率多时相遥感数据进行作物分类,采用决策树分类、神经网络分类两种方法的精度分别为72.0729%、87.3%。利用决策树分类的效果最优。
Abstract
Taking the entire county of Yongqing County, Langfang City, Hebei Province as the research area, the GF1-WFV 16-meter resolution image was used as the data source, and the image data covering multiple phases of the crop growth period was selected to construct the nonralized difference vegetation index (NDVI) time series of the crop. By analyzing the NDVI curve of the study area, it was found that the NDVI time series constructed with the data could describe the growth characteristics of crops in the study area, reflect the phenological differences of different crops in the region, and effectively distinguish the local planting patterns. The decision tree is constructed by selecting characteristic parameters such as the occurrence time of the maximum value, minimum value, peak, peak number and threshold value on the NDVI curve. According to the phenological calendar of the study area and the investigation of the local planting structure, the optimal time phase image was used to classify and extract one or several crops. Decision tree classification and neural network were used respectively, and accuracy verification was carried out to obtain the best crop classification method by comprehensive comparison. The results show that in the county research area of Yongqing County, when the GF1-WFV 16-meter resolution multi-temporal remote sensing data is used for crop classification, the accuracy of decision tree classification and neural network classification are 72.0729% and 87.3%, respectively. The optimal classification is obtained by using decision tree.

李龙, 李旭青, 吴伶, 杨秀峰, 孙鹏飞. 基于决策树和神经网络的农作物分类研究——以廊坊市为例[J]. 红外, 2019, 40(3): 24. LI Long, LI Xu-qing, WU Ling, YANG Xiu-feng, SUN Peng-fei. Crop Classification Research Based on Decision Tree and Neural Network——Take Langfang City as an Example[J]. INFRARED, 2019, 40(3): 24.

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!