农作物冠层光谱信息的施肥管理分区研究

随着地面遥感技术的不断发展，越来越多的农作物冠层光谱检测传感器被应用到了农业生产，其中应用较为广泛的就是Greenseeker植物光谱检测仪，利用Greenseeker植物光谱检测仪可以获取农作物冠层光谱信息归一化植被指数（NDVI）数据，从而能够进行农作物的施肥管理分区的划分，依据划分好的施肥管理分区可以实现有针对性的变量施肥。模糊c-均值（FCM）算法是划分农作物施肥管理分区常用的算法，但是模糊c-均值算法具有一定的局限性，就是在计算过程中随着NDVI数据量的增加会不断进行数据的迭代计算，从而会影响施肥管理分区划分的速度。在模糊c-均值算法的基础上提出一种基于模型的模糊c-均值（MFCM）算法，基于模型的模糊c-均值算法在划分农作物施肥管理分区过程中不必在每获取一组数据时就对全部数据进行迭代计算，可有效提高划分施肥管理分区的速度。通过搭建的农作物冠层光谱信息采集平台获取大豆和玉米的NDVI数据，利用基于模型的模糊c-均值算法划分大豆和玉米的施肥管理分区，使用分区评价指标轮廓系数（SC）和调整兰德指数（ARI）评价划分施肥管理分区的效果。结果表明，随着获取的NDVI数据量的不断增加，基于模型的模糊c-均值算法相比于模糊c-均值算法能够更快的划分施肥管理分区，在划分大豆施肥管理分区上，基于模型的模糊c-均值算法快0.02～0.15 s；在划分玉米施肥管理分区上，基于模型的模糊c-均值算法快0.07～0.51 s。通过计算评价划分施肥管理分区效果的指标轮廓系数和调整兰德指数发现，在不同NDVI数据量的情况下进行划分施肥管理分区，轮廓系数的值最大相差为0.022，说明两种算法划分施肥管理分区的效果相差不大；调整兰德指数的值对数据的波动变化比较敏感，在NDVI数据量达到6 000后能够维持在0.7以上，但当NDVI数据波动变化较大时会出现一定的下降。

Fertilization Management Zoning Based on Crop Canopy Spectral Information

With the continuous development of ground remote sensing technology, more and more crop canopy spectral sensors are applied to agricultural production, among which the Greenseeker plant spectral detector is widely used. Greenseeker can obtain crop canopy spectral information, normalized vegetation index (NDVI) data and divide fertilization management zoning. Targeted variable rate fertilization can be realized according to fertilization management zoning. The fuzzy c-means (FCM) algorithm is common for dividing fertilization management zoning, but the FCM algorithm has certain limitations. In the calculation process, the iterative calculation will be carried out continuously with the increase of data, which will affect the speed of fertilization management zoning. Based on the FCM algorithm, a model-based fuzzy c-means (MFCM) algorithm is proposed. In dividing the fertilization management partition, this algorithm does not have to iteratively calculate all the data every time a group of data is obtained, which can improve the speed of dividing the fertilization management partition. The NDVI data of soybean and maize were obtained through the established crop canopy spectral information collection platform. The fertilization management zoning was divided by the MFCM algorithm, and the division effect was evaluated by evaluation index contour coefficient (SC) and adjusted rand index (ARI). The results show that with the increased NDVI data, the MFCM algorithm can partition fertilization management partition faster than the FCM algorithm. The MFCM algorithm is 0.02～0.15 seconds faster; the MFCM algorithm is 0.07～0.51 seconds faster in dividing maize fertilization management zoning. By calculating the indexes SC and ARI to evaluate the effect of dividing fertilization management zoning, it is found that when dividing different NDVI data, the maximum difference of SC value is 0.022, indicating that the effect of dividing fertilization management zoning by the two algorithms is not different; The ARI value is sensitive to data changes. It can be maintained above 0.7 after the NDVI data volume reaches 6 000, but it will decrease significantly when the NDVI data changes.