正态分布下最小错误率的变量喷施贝叶斯决策

传统农田除草采用田间统一定量均匀喷洒,导致了除草剂浪费和环境污染问题。智能变量喷施能够保护环境和提高作物产量,是促进农业可持续发展战略的重要途径。为此,对经典的杂草监测参数进行改进并提出了正态分布下最小错误率的贝叶斯决策以实现精确变量喷施。首先对农田图像进行灰度化、二值化及去噪等预处理;然后依据作物行中心线对农田图像进行网格单元的划分,并在网格单元格内提取改进的杂草监测参数;最后将贝叶斯决策分为两个阶段:线下阶段利用改进的杂草监测参数数据库计算正态分布参数,线上阶段根据改进的杂草监测参数实现正态分布下最小错误率的贝叶斯决策,从而为变量喷施提供决策依据。实验结果表明:正态分布下最小错误率的贝叶斯决策正确率可达92%,与BP算法和SVM算法相比决策正确率相对较高。

Variable Spraying Based on the Minimum Error Ratio of Bayes Decision under Normal Distribution

Traditional farmland spraying is united quantitative and evenly, and this cause the waste of herbicide and envi-ronment pollution.Intelligent variables spraying, which not only can protect environment but also increase crop output, is the crucial way to promote sustainable agriculture development.In this paper, first modified the classic weed infestation rate( WIR) and then an accurate variables spraying based on the minimum error Bayes decision under normal distribution is presented.Firstly, farmland images are pre-processing using graying, binary and de-noising.Secondly, grid unit of farmland images are divided according to the centerline of crop rows and then compute the modified weed infestation rate ( MWIR) in the grid unit.Finally, bayesian decision is divided into two stages.Normal distribution parameters are com-puted base on database of MWIR in the offline stage, and Bayes online decision based on minimum error according to MWIR under normal distribution, which provide basis for decision making of intelligent variables spraying.The experi-mental results showed that the accuracy of this algorithm is as high as 92%, which exceeded BP algorithm and SVM algo-rithm.