基于LVQ和THz时域光谱的玉米品种分类鉴别研究

Research on Classification and Identification of Maize Varieties Based on LVQ and THz Time Domain Spectra

农作物的产量和品质与其自身的品种密切相关,因此品种鉴别对于农业生产和安全具有极为重要的意义。研究提出将太赫兹时域光谱(THz-TDS)与神经网络学习矢量量化(LVQ)相结合的方法用于四类玉米品种鉴别。实验选取120粒玉米样本的衰减全反射(ATR)吸收系数0～70 cm^(-1)谱区作为LVQ网络输入,4个品种作为网络输出,随机划分测试集与训练集的样本比例为1:1、1:2和1:5时,测试集识别率分别为80%、82. 5%、95%;实验选取ATR吸收系数0～275 cm^(-1)全谱区作为LVQ网络输入时,测试集识别率分别为93. 3%、97. 5%、100%。实验结果表明采用太赫兹光谱结合LVQ方法能有效鉴别玉米品种,该方法对于农作物品种快速鉴别具有一定的借鉴性。

The yield and quality of crops are closely related to their own varieties, so the identification of varieties was of great importance to agricultural production and safety. In this paper, the method of combining Terahertz time domain spectroscopy(THz-TDS) with neural network learning vector quantization(LVQ) was proposed for the identification of four types of maize varieties. The 0-70 cm-1 spectral region of the Attenuated Total Reflection(ATR) absorption coefficient of 120 maize samples was selected as the LVQ network input, and four varieties were used as the network output. And the ratio of the sample set to the training set was 1: 1, 1: 2 and 1:5. The recognition rate of the test set was 80%, 82. 5% and 95%. While the full spectrum with 0-275 cm -1 of the ATR absorption coefficient was selected as the LVQ network. The recognition rate of the test set was 93. 3%, 97.5% and 100%. The experimental results showed that the maize varieties can be effectively identified by terahertz spectroscopy combined with LVQ method, which was capable of playing a reference role in the rapid identification of crop varieties.