基于FCM的绿洲农田养分管理分区研究

 【目的】以天山北麓绿洲农耕区——新疆农八师148团农田为研究对象，以193个耕层土壤（0～30 cm）有机质、碱解氮、速效磷和速效钾含量的分析数据为变量进行农田土壤养分精确管理分区研究。【方法】模糊c-均值聚类法被用来进行分区，以棉田产量为外部变量，采用FPI、c-φ多次组合法及基于外部变量的多元回归法来确定适宜的模糊控制参数。【结果】研究区最佳分区数为4，模糊指数为1.6。各管理分区土壤养分的变异系数都较分区前全研究区有所减小，而分区间土壤养分差异显著。研究区的平均混乱度指数为0.19，不同模糊类别交叠程度小，地理空间上土壤的隶属关系相对明确。【结论】通过选取适宜的外部变量，模糊c-均值聚类法可以较好地进行管理分区划分，分区结果可以作为变量施肥的单独作业单元进行耕作管理。

Definition of Management Zones of Soil Nutrients Based on FCM Algorithm in Oasis Field

【Objective】OM, AN, AP, AK determined in 193 topsoil (0-30cm) samples were collected from the cotton field in the 148th regiment of the 7th Agricultural Division in Xinjiang, the study of management zones was achieved by the use of fuzzy c-mean algorithm (FCM).【Method】In this paper fuzzy c-means clustering algorithm was used to delineate management zones and the cotton yield was chosen as the external variable, FPI (Fuzziness Performance Index) and the derivative of the objective function with respect to fuzziness exponent and the adjusted coefficient of determination of the external variable were used to select the most suitable number of classes and fuzziness exponent. 【Result】The optimal number of management zones for the study area was 4. Management zone 1 presented the highest nutrient level, whereas management zone 2 was the lowest. Meanwhile, the mean confusion index of study area was 0.1875.【Conclusion】It was demonstrated that FCM approach combined with geostatistical techniques could produce a desirable prediction results which has more information compared with conventional direct kriging methods. On the other hand, the external variable can help to select the most suitable number of classes and fuzziness exponent.The results revealed that the defined management zones not only direct soil-sampling design, but can be used for fertilizer recommendation to manage soil properties more efficiently in semiarid cotton system of Xinjiang.