School of Engineering, University of Warwick, Coventry CV4 7AL, UK; Department of Physics Education, Universitas Negeri Yogyakarta, Yogyakarta, 55281 Indonesia.
Department of Biology Education, Universitas Negeri Yogyakarta, Yogyakarta, 55281 Indonesia.
Spectrochim Acta A Mol Biomol Spectrosc. 2022 Oct 5;278:121339. doi: 10.1016/j.saa.2022.121339. Epub 2022 May 4.
Pre-processing is a crucial step in analyzing spectra from Fourier transform infrared (FTIR) spectroscopy because it can reduce unwanted noise and enhance system performance. Here, we present the results of pre-processing technique optimization to facilitate the detection of pepper yellow leaf curl virus (PYLCV)-infected chilli plants using FTIR spectroscopy. Optimization of a range of pre-processing techniques was undertaken, namely baseline correction, normalization (standard normal variate, vector, and min-max), and de-noising (Savitzky-Golay (SG) smoothing, 1st and 2 derivatives). The pre-processing was applied to the mid-infrared spectral range (4000 - 400 cm) and the biofingerprint region (1800 - 900 cm) then the discrete wavelet transform (DWT) was used for dimension reduction. The pre-processed data were then used as an input for classification using a multilayer perceptron neural network, a support vector machine, and linear discriminant analysis. The pre-processing method with the highest classification model accuracy was selected for the further use in the processing. It was seen that only the SG 1st derivative method applied to both wavenumber ranges could produce 100% accuracy. This result was supported by principal component analysis clustering. Thus, we have demonstrated that by using the right pre-processing technique, classification success can be increased, and the process simplified by optimization and minimization of the technique used.
预处理是分析傅里叶变换红外（FTIR）光谱的关键步骤，因为它可以减少不必要的噪声并提高系统性能。在这里，我们介绍了预处理技术优化的结果，以促进使用 FTIR 光谱检测感染胡椒黄花叶病毒（PYLCV）的辣椒植物。对一系列预处理技术进行了优化，包括基线校正、归一化（标准正态变量、向量和最小-最大值）和去噪（Savitzky-Golay（SG）平滑、一阶和二阶导数）。预处理应用于中红外光谱范围（4000-400 cm）和生物指纹区域（1800-900 cm），然后使用离散小波变换（DWT）进行降维。预处理后的数据然后用作分类的输入，使用多层感知机神经网络、支持向量机和线性判别分析。选择具有最高分类模型准确性的预处理方法用于进一步处理。结果表明，仅在两个波数范围内应用 SG 一阶导数方法即可达到 100%的准确性。主成分分析聚类支持了这一结果。因此，我们已经证明，通过使用正确的预处理技术，可以提高分类成功率，并通过优化和最小化所使用的技术简化该过程。
