Determination of methylene violet concentration using classification algorithms

The dyestuffs used in the industry are harmful to human health and the environment. One of the most widely used of these dyestuffs is methylene violet (MV). Detection of such substances and determination of their concentrations are significant for the development of purification processes. Although there are many spectrophotometric methods and sensitive devices for the determination of dyestuff in aqueous solutions, these devices can be only used in the lab, cannot be employed beyond the laboratory, and are costly. Therefore, artificial intelligence-based systems can be preferred outside the laboratory and in case of urgent use in terms of accessibility and practicality. In this study, we photographed images of MV solutions in concentrations of 0.1–1–5–25–50 ppm. Image features are extracted using a deep learning architecture for each image found in the data set created according to different concentration values. Afterward, Linear Discriminant, Linear Support Vector Machine (SVM), Cubic SVM, Quadratic SVM, and Subspace Discriminant Ensemble classifiers were trained by employing the extracted image features. Using these trained classifiers, we designed a user-friendly Graphical User Interface (GUI) application that can predict the concentrations of MV solutions to provide advantages in terms of time savings, cost, and practicality. The results show that the most successful classifier used in the study is the Subspace Discriminant Ensemble.