Development of Spatial-Temporal Model of Soil-Plant System and Farm Management System

1)研制了拖拉机式土壤光谱传感器，实现了15 ~ 40 cm深度土壤反射率的连续实时检测和现场位置记录。它提供400 ~ 1700 nm波长的土壤光谱反射率，从可见光到近红外范围，可用于预测土壤水分、土壤有机质含量、硝酸盐氮含量、电导率和ph。所提出的土壤传感器将有可能提高土壤参数测绘的准确性，同时为工程节省时间和劳动力。2)提出了一种基于日常天气变化风险管理的农活调度算法。风险是通过成本来评估的，然后在优化中最小化成本。采用遗传算法进行优化，以获得柔性。3)采用改进的Lotka-Volterra模型处理密度依赖和对称竞争。在模型中引入距离函数，并在此基础上进行参数研究。讨论了logistic模型的可行性。获得了三叶草与杂草竞争过程的实验数据，并基于非对称模型进行了非线性回归。对称模型与实验数据吻合较好。4)我们将微选图像作为随机纹理，作为特征提取器。从分类的45个特征元素中选取20个特征元素构成线性判别函数。98%的训练区分类正确，88%的测试区从混合植被的照片中与人类的判断一致。

1) A tractor-mounted spectroscopic soil sensor developed has enabled to real-time detection of the sub-soil reflectance continuously at depths of 15 to 40 cm as well as to recording the location in the field. It provides spectral reflectance of soil over 400 to 1700 nm wavelengths, a visible to NIR range, which are available for predicting soil moisture, soil organic matter content, nitrate nitrogen content, electric conductivity and pH. The proposed soil sensor will make it possible to increase the accuracy in soil parameters mapping, in addition to time and labor saving for the works.2) An algorithm for farm work scheduling was developed based on managing the risk of daily weather variation. The risk was evaluated by costs which were then minimized in the optimization. Genetic algorithms were utilized for optimization in order to attain flexibility.3) Density dependence and symmetrical competition were dealt with by the modified Lotka-Volterra model. Distance function was incorporated into the model and parameter studies based on the model were conducted. Feasibility of the logistic model was discussed. Experimental data for clover-weed competition process were obtained and nonlinear regression was conducted based on the non-symmetrical model. The symmetrical model showed good agreement with experimental data.4) We treat a picture of micro-selected to constitute a random texture and adopted as a feature extractor. 20 feature elements among 45 are classified are selected to constitute a linear discriminative function. 98% of training sections are classified correctly and 88% of test sections from photographs of mixed vegetation of both stocks agreed with human judgment.