基于前馈神经网络的变量施肥养分图的研究

Study on Feedforward ANN Model on Variable Rate Fertility Map

通过插值方法合理利用土壤采样点的养分值估算非采样点的养分值,对于变量施肥决策和实施具有重要的意义。由于大采样间距插值地图的精确性无法得到保证,难以客观地描述农田的养分分布情况,因此提出一种不依赖于地理统计学推论的前馈神经网络模型,对大间距采样点进行插值。实验研究表明,与目前的插值方法相比,该模型有较好的准确性和可靠性。

It is an important means to utilize the methods of interpolation to estimate non-sample values for variable rate fertilizer decision-making .The methods of interPolation is introduced .Yet ,great distances between sample values make the accuracy of interpolation maps uncertain. In this paper, a kind of feedforward ANN model was implemented in agronomic area ,which was not independent of geography statics deduction ,the test result of experiment showed that model gave accurate results and worked steadily.