基于光谱时间序列拟合的中国南方水稻遥感识别及面积估算方法研究

发布时间：2018-04-15 16:16

本文选题：水稻 + 遥感分类　；参考：《东华理工大学》2017年硕士论文

【摘要】：粮食安全对于一个国家的发展至关重要,农作物种植面积是保障粮食安全的基础,因此农作物种植面积估算非常重要。目前遥感技术在识别农作物和面积估算领域应用非常广泛。多云多雾现象是农作物遥感分类经常遇到的问题,影响识别精度。目前有学者通过云检测方法将有云的区域识别并剔除出去,用其他多期影像补充,虽然解决了云的问题,但是一般情况下多期影像并不是识别农作物的最佳时相,依然无法保证分类精度。本文提出一种基于光谱时间序列拟合的水稻遥感识别方法,即解决了多云多雾现象影响水稻遥感识别的问题,又能保证精度。本文水稻识别方法是利用时间序列的GF-1号遥感影像提取中稻、晚稻的近红外波段反射率(NIR)、红光波段反射率(R)、归一化植被指数(NDVI)特征,拟合中稻、晚稻光谱(近红外、红光)和归一化植被指数时间序列特征曲线,分析参与时间序列特征曲线拟合的多时相影像近红外波段、红光波段、NDVI值落在拟合中稻、晚稻近红外波段、红光波段和NDVI时间序列曲线两侧的敏感性区域的比例,该敏感区域也可以视为水稻识别区域,只有达到一定的比例才能视为某类水稻作物。在此情形下,需要综合三种情况进行集中投票决定其最终分类结果。本文统计出外业调查样方内水稻真实种植面积,以一部分样方作为修正样方,剩余样方用于面积估算精度验证,建立GF-1号WFV分类水稻面积与样方水稻面积之间的线性关系,通过线性回归模型计算出水稻面积修正参数,利用线性修正模型估算出验证样方内水稻面积,与验证样方内真实水稻面积进行对比,验证精度,最后对整个研究区GF-1号WFV数据水稻的分类统计结果进行修正,估算出研究区水稻种植面积。研究表明:本文识别方法可以在多云雾地区对中稻和晚稻精确识别,中稻和晚稻用户精度分别可达95.97%和95.95%,总体精度为95.76%,kappa系数为0.9335,结果较为理想,表明了NIR、R和NDVI时间序列曲线拟合的有效性,以及拟合曲线水稻识别区域设置的合理性,解决了多云多雾对水稻遥感识别的问题。最后利用光谱时间序列拟合水稻识别方法获得研究区宏观监测水稻种植面积,运用水稻面积修正模型对研究区水稻进行修正,有效提高了面积估算精度。
[Abstract]:Food security is very important for the development of a country, crop planting area is the basis of ensuring food security, so the estimation of crop planting area is very important.At present, remote sensing technology is widely used in the field of crop identification and area estimation.Cloud and fog phenomenon is a common problem in crop remote sensing classification, which affects the recognition accuracy.At present, some scholars use cloud detection methods to identify and eliminate areas with clouds and supplement them with other multi-phase images. Although the cloud problem is solved, multi-phase images are not the best time to identify crops in general.Classification accuracy is still not guaranteed.In this paper, a rice remote sensing recognition method based on spectral time series fitting is proposed, which solves the problem that the cloud and fog phenomenon affects rice remote sensing recognition and ensures the precision.In this paper, the method of rice identification is to extract the characteristics of middle rice, near infrared band reflectance (NIR), red light band reflectance (RN), normalized vegetation index (NDVI) of middle rice and late rice by using GF-1 remote sensing image of time series, and to fit the spectra of middle rice and late rice (near infrared).Red light) and normalized vegetation index time series characteristic curves were analyzed, and the near infrared bands of multiphase images involved in fitting time series characteristic curves were analyzed. The NDVI values of red light bands fell in the near infrared bands of fitting middle rice and late rice.The ratio of the sensitive region on both sides of the red light band and the NDVI time series curve can also be regarded as the rice identification area. Only when a certain proportion is reached can the sensitive region be regarded as a certain kind of rice crop.In this case, the final classification results need to be determined by centralized voting in three cases.In this paper, the real planting area of rice in field survey was counted, and a part of the sample was used as the modified sample, the remaining sample was used to verify the accuracy of the area estimation, and the linear relationship between the rice area classified by GF-1 WFV and the area of the sample rice was established.The rice area correction parameters were calculated by linear regression model, and the rice area in the validation sample was estimated by linear correction model, and compared with the real rice area in the validation sample, the accuracy was verified.Finally, the classification and statistical results of GF-1 WFV data in the whole research area were revised, and the rice planting area was estimated.The results show that this method can be used to identify middle and late rice precisely in cloudy and foggy areas. The user accuracy of middle rice and late rice is 95.97% and 95.95%, respectively. The overall accuracy is 95.76% and the coefficient of kappa is 0.933 5. The results are satisfactory.The results show the validity of curve fitting of NIR R and NDVI time series, and the rationality of rice identification region setting of fitting curve, which solves the problem of rice remote sensing recognition by cloudy and fog.Finally, using the spectral time series fitting rice identification method to obtain the macroscopic monitoring of rice planting area in the study area, using the rice area correction model to modify rice in the study area, effectively improve the precision of area estimation.
【学位授予单位】：东华理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S511;S127

【参考文献】