基于遥感影像的张掖灌区作物种植结构提取研究

Crop Planting Structure Extraction in Zhangye Irrigation Area Based on Remote Sensing Images

基于遥感影像的作物种植结构提取在实践中得到广泛应用,其中分类特征和样本数量选取是影响提取精度的关键。为了探讨不同分类特征和样本数量对作物种植结构提取精度的影响,以甘肃张掖灌区为研究区,采用监督分类中支持向量机方法,研究了光谱与时序NDVI两种分类特征在不同样本数量条件下的作物种植结构提取精度。结果表明:①随着样本数量的增加,识别的作物种植结构空间分布准确性逐渐增加直至稳定状态。②基于时序NDVI特征提取的玉米面积平均误差为2.82%,平均总体分类精度为84.8%,平均Kappa系数为0.81;其精度优于基于光谱特征提取结果。③研究区每10 km2的样本数量为3~4个时,样本能够保持最佳的训练效果。研究成果可为提高作物种植结构提取精度提供重要参考。

Crop planting structure extraction based on remote sensing images has been widely used in practice,in which the selection of clas‐sification features and samples are the key factors affecting the extraction accuracy.In order to explore the influence of different classification features and samples on the extraction accuracy of crop planting structure,with the study area in Zhangye Irrigation Area in Gansu Province,this paper uses support vector machine supervised classification method of studying the extraction accuracy of crop planting structure of spec‐tral and temporal NDVI classification features under different samples.The results show that:①With the increase in the number of sam‐ples,the accuracy of the spatial distribution of the identified crop planting structure gradually increase to a stable state.②The average error of corn area extracted by temporal NDVI is 2.82%,the average overall classification accuracy is 84.8%,and the average Kappa coefficient is 0.81.The accuracy of crop planting structure based on temporal NDVI feature extraction is better than that of spectral feature extraction.③When the number of samples per 10km2 in the study area is 3~4,the samples can maintain the best training effect.The research results can provide an important reference for improving the extraction accuracy of crop planting structure.