基于遗传算法的近红外光谱定性分析特征波长提取方法研究

Study on Characteristic Wavelength Extraction Method for Near Infrared Spectroscopy Identification Based on Genetic Algorithm

近红外光谱分析技术虽在多领域获得广泛应用,但应用时仍以实验室仪器为主,目前光谱仪存在体积大、功耗高、价格贵等问题,有能力购买与使用此类仪器的主要是高校、科研院所、大型企业等,常用的基于傅里叶变换或光栅原理的光谱仪价格通常高达几十万元,超出中小企业、普通百姓的经济承受能力,因此近红外光谱仪的进一步推广应用仍有难度。降低仪器造价并实现微型化,是推广近红外光谱技术应用的一个重要方向,近红外光谱仪小型化的努力方向有CT正交型光栅技术以及微机电系统技术,但这两种技术方案对光谱仪体积缩小幅度有限,仍存在价格高、内部有移动部件等问题,难以做到真正微型化。据光谱仪的工作原理可知,其价格高低及微型化难度与仪器所能检测波段以及分辨率密切相关,以线性渐变滤光片与InGaAs探测器为例,分辨率越高,检测的波长点越多,其价格越高,制造难度越大。针对某一特定的定性分析任务,若能从大量波长点中挑选出少量特征波长点,并利用挑选得到的少量特征波长点完成对被测样本的定性分析任务,则可降低仪器制造成本,并降低光谱仪微型化难度,从而有利于近红外光谱分析技术的推广与应用。以玉米单倍体和多倍体籽粒作为研究对象,针对两类籽粒分类任务,分多天以漫透射方式采集被研究对象的近红外光谱,按时间顺序将所采数据分为5个数据集,对第1个数据集使用遗传算法提取出10个特征波长点,再将提取得到10个特征波长点,用于剩余4个数据集的单倍体、二倍体鉴别,以检验方法的有效性。实验结果表明使用10个特征波长点能够获得与全光谱基本一致的鉴别效果,说明使用少量特征波长点上的吸光度值也能够有效鉴别单倍体,可为其他领域某特定任务开发低成本便携式微型近红外光谱仪提供借鉴。

At present,although the near-infrared(NIR)spectroscopy analysis technology has been widely used in many fields,it is mainly used as laboratory instruments,and the spectrometer used in the laboratory has the problems of large volume,high power consumption and high price.The main units that can purchase and use the NIR spectrometer are universities,scientific research institutes,large and medium-sized enterprises,etc.The price of a spectrometer based on the Fourier or grating principle is usually as high as several hundred thousand Yuan,which is beyond the affordability of small and medium-sized enterprises and ordinary people.Therefore,the application of NIR spectrometer is far away from ordinary people’s daily life.The high price and large volume of near-infrared spectrometers restrict the large-scale application of near-infrared spectroscopy analysis technology.The reason is that the near-infrared spectrometer itself is expensive and the volume has not yet been portable and miniaturized.Reducing the cost of the NIR spectrometer and miniaturizing the spectrometer is an important direction to promote NIR spectroscopy technology.The efforts of miniaturization of NIR spectrometer include CT orthogonal grating technology and micro electro mechanical system technology.However,the volume reduction of the spectrometer by these two technical solutions is limited,and there are still some problems,such as high price,internal moving parts and real hard miniaturization.For a specific qualitative analysis task,a small number of characteristic wavelength points are selected from full spectra and used to build models which can recognize testes samples.The method mentioned above can reduce the cost of instrument manufacturing and difficulty of spectrometer miniaturization,and it is also conducive to the large-scale promotion and application of NIR analysis technology.Near infrared spectra of Haploid and diploid maize seeds are collected by diffuse transmission method in several days.The collected data are divided into five data sets in chronological order.For the first data set,10 characteristic wavelength points are extracted by genetic algorithm,and then 10 characteristic wavelength points are extracted for the remaining four data sets In order to test the validity of the method,the haploid and diploid identification was carried out.The experimental results show that using 10 characteristic wavelength points can obtain the identification effect,which is consistent with the full spectrum,indicating that using a small number of characteristic wavelength points can also effectively identify haploids,which can provide a reference for the development of low-cost portable NIR spectrometer for a specific task in other fields.