马铃薯红外光谱数据库系统关键算法研究

发布时间：2019-06-13 09:31

【摘要】：红外光谱以其稳定性高、无需对样品做化学处理、图谱信息含量丰富等优点,被广泛应用于化学品分析、物质品种预测、品质质量鉴定等诸多领域。红外光谱数据库系统能帮助建立稳定、快捷的样品种类预测模型、质量分析模型、特征分析模型等,使研究人员可以更加全面地掌握样品的信息。在红外光谱数据库系统中,准确、高效的光谱分类匹配算法是整个系统有效运行的关键保证,因此,研究光谱分类匹配算法对推广红外光谱数据库系统有促进作用。现有的光谱匹配算法多从欧氏距离测度或曲线相似性方面衡量光谱之间的相似程度,而无法将两方面因素综合,并且随着类别中心总数的增加,现有算法的准确率已无法满足光谱数据库系统的要求。因此本文以马铃薯作为研究对象,研究了红外光谱数据库系统中的关键算法。首先,针对传统光谱特征峰识别算法需要对光谱做多次扫描、对矮小峰及肩峰识别能力较弱的现象,提出了一种只需一次扫描的基于动态峰形因子的特征峰识别算法。实验表明,该算法能够准确地识别出光谱曲线中的所有有效特征峰,对肩峰及矮小峰也有一定的识别能力。其次,根据汉明距离及光谱差曲线的概念提出了动态光谱距离算法,该算法综合考虑了光谱曲线的波形因素及绝对差异因素,实现了不同品种马铃薯的精确鉴别,实验结果表明,该算法的平均准确率达到92.85%,高于传统的欧氏距离、光谱角等算法。最后,针对类别中心总数不断增多时,光谱分类算法准确率下降的问题,提出了基于虚拟竞争自组织自增长特征映射神经网络VC-TGSOM的光谱分类算法。实验表明,VC-TGSOM网络的准确率不会随着类别中心总数的增加而下降。
[Abstract]:Infrared spectroscopy has been widely used in many fields, such as chemical analysis, material variety prediction, quality identification and so on, because of its high stability, no need for chemical treatment, rich atlas information and so on. Infrared spectrum database system can help to establish stable and fast sample type prediction model, quality analysis model, feature analysis model and so on, so that researchers can master the sample information more comprehensively. In infrared spectral database system, accurate and efficient spectral classification and matching algorithm is the key to the effective operation of the whole system. Therefore, the study of spectral classification and matching algorithm can promote the promotion of infrared spectral database system. Most of the existing spectral matching algorithms measure the similarity between spectra from the aspect of Euclidean distance measure or curve similarity, but can not synthesize the two factors, and with the increase of the total number of category centers, the accuracy of the existing algorithms can no longer meet the requirements of spectral database system. Therefore, taking potato as the research object, the key algorithm of infrared spectrum database system is studied in this paper. Firstly, aiming at the fact that the traditional spectral feature peak recognition algorithm needs to scan the spectrum many times, and the recognition ability of dwarf peak and shoulder peak is weak, a feature peak recognition algorithm based on dynamic peak shape factor is proposed, which only needs one scan. The experimental results show that the algorithm can accurately identify all the effective characteristic peaks in the spectral curve, and also has a certain ability to recognize shoulder peaks and dwarf peaks. Secondly, according to the concepts of hamming distance and spectral difference curve, a dynamic spectral distance algorithm is proposed. The algorithm takes into account the waveform and absolute difference factors of spectral curve, and realizes the accurate identification of different varieties of potato. The experimental results show that the average accuracy of the algorithm is 92.85%, which is higher than the traditional Euclidean distance, spectral angle and so on. Finally, in order to solve the problem that the accuracy of spectral classification algorithm decreases when the total number of category centers increases, a spectral classification algorithm based on virtual competitive self-organizing self-growing feature mapping neural network VC-TGSOM is proposed. The experimental results show that the accuracy of VC-TGSOM network does not decrease with the increase of the total number of category centers.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：S532;TP311.13

【参考文献】