Spectral classification by generative adversarial linear discriminant analysis

Vibrational Spectroscopy Can Be Vulnerable to Adversarial Attacks

Nondestructive detection methods based on vibrational spectroscopy have been widely used in many critical applications in a variety of fields such as the chemical industry, pharmacy, national defense, security, and so on. As these methods/applications rely on machine learning models for data analysis, studying the threats associated with adversarial examples in vibrational spectroscopy and defenses against them is of great importance. In this paper, we propose a novel adversarial method to attack vibrational spectroscopy, named SynPat, where synthetic peaks produced by a physical model are placed at key locations to form adversarial perturbations. Our new attack generates perturbations that successfully deceive machine learning models for Raman and infrared spectrum analysis while they blend much better into the spectra and hence are unnoticeable to human operators, unlike the existing state-of-the-art adversarial attacking methods, e.g., images and audio. We verified the superiority of the proposed SynPat by an imperceptibility test conducted by human experts and of defense experiments by an AI detector. To the best of our knowledge, this is a first thorough study on the robustness of vibrational spectroscopic techniques against adversarial samples and defense mechanisms. Our extensive experiments show that machine learning models for vibrational spectroscopy, including conventional and deep models for Raman or infrared classification and regression, are all vulnerable to adversarial perturbations and thus may pose serious security threats to our society.

Research on the chemical oxygen demand spectral inversion model in water based on IPLS-GAN-SVM hybrid algorithm

Spectral collinearity and limited spectral datasets are the problems influencing Chemical Oxygen Demand (COD) modeling. To address the first problem and obtain optimal modeling range, the spectra are preprocessed using six methods including Standard Normal Variate, Savitzky-Golay Smoothing Filtering (SG) etc. Subsequently, the 190-350 nm spectral range is divided into 10 subintervals, and Interval Partial Least Squares (IPLS) is used to perform PLS modeling on each interval. The results indicate that it is best modeled in the 7th range (238~253 nm). The values of Mean Square Error (MSE), Mean Absolute Error (MAE) and R2score of the model without pretreatment are 1.6489, 1.0661, and 0.9942. After pretreatment, the SG is better than others, with MSE and MAE decreasing to 1.4727, 1.0318 and R2score improving to 0.9944. Using the optimal model, the predicted COD for three samples are 10.87 mg/L, 14.88 mg/L, and 19.29 mg/L. To address the problem of the small dataset, using Generative Adversarial Networks for data augmentation, three datasets are obtained for Support Vector Machine (SVM) modeling. The results indicate that, compared to the original dataset, the SVM's MSE and MAE have decreased, while its accuracy has improved by 2.88%, 11.53%, and 11.53%, and the R2score has improved by 18.07%, 17.40%, and 18.74%.