基于遗传多层前馈神经网络的大豆脂肪酸含量近红外光谱检测

文章提出了一种利用遗传多层前馈神经网络建立数学模型的方法,建立起化学测定值与近红外光谱数据之间的定量关系。把得到的近红外光谱数据作为网络的输入,把用化学法测定的5种脂肪酸含量作为网络的输出,再利用遗传算法训练多层前馈神经网络的权值,建立大豆脂肪酸的神经网络检测模型,探索出一种能够准确、高效地完成近红外光谱检测的神经网络模型,文中设计了一种用遗传算法训练的多层前馈神经网络。通过试验证明,用遗传算法优化人工神经网络的权重,获得高于单纯用人工神经网络训练的结果。大豆5种脂肪酸的相关系数都可达到0.9左右,能够满足大豆育种的初步检测。

Near infrared spectrum detection of soybean fatty acid content based on genetic multilevel forward neural network

This paper represented a way to build mathematical model on genetic multilevel forward neural network,built the relationship between chemistry measurement values and near infrared spectrum datum.The near infrared spectrum data was input in this network,five kinds of fatty acid contents,which measured by chemistry method,were output.Then,trained the weight of multilevel forward neural network by genetic algorithms,built the soybean fatty acids neural network detection model,and explored the network model which can realize near infrared spectrum detection exactly and efficiently.A multilevel forward neural network trained by genetic algorithms was designed.Test showed that the relative coefficient in five fatty acids of soybean can be round about 0.9,and can satisfy initial detection of soybean breeding.