基于神经网络的树形识别系统研究

常规的施药方法,由于不考虑树与树之间的间距、树冠的大小与形状,大部分农药未能发挥效用,流失非常严重,喷撒出去的农药只有极少部分能到达要防治的靶标上,从而造成了巨大的农药浪费,且带来了农药安全、环境污染等诸多问题。笔者在分析农药的喷药量与树形之间关系的基础上,对原有的基于BP神经网络的树形识别系统进行了改进设计,重点对该网络的输入层、隐含层及输出层进行了理论分析和试验研究,最终确定了由15个节点组成输入层,分别代表树冠的形状特征和树木图像的分形维数特征;6个节点组成隐含层,6个节点组成输出层,分别代表要判断树形的种类。对16个样本实时处理的试验表明,除了1个样本因未能分割外,其他样本均可识别,而且稳定性有了很大提高,适合于农药的精确施用。

Improved Tree Crown Recognition System Based Artificial Neural Network

Pesticides have been applied homogeneously no matter whether there were somespaces among sprayedtarget trees .Theindividual differences of tree crowns were consideredas a key problem. The i mproved tree crown recognition systembased on BP artificial neuralnetworks was used to increase esti mated precision of target trees in this paper .It also em-phasized the experi mental and theoretical studies on determination of nodes in layers ofANN. There are 15 nodes in the input layer ,6 nodes in the hiddenlayer and 6 nodes in theoutput layer .The experi mentsinthe real ti me showedthat thei mprovedtree crownrecogni-tion system was more practical than previous one .