ANFIS-Based Determination of pH Level of Liquid Raw Materials with Image Processing

Abstract

Determination of pH level in liquid materials are important phenomenon, especially to determine the quality of the raw material, and some mechanical features like abrasive characteristics, and so on. Today's conventional methods are based on spectroscopic or non-spectroscopic methods, such as refractometric, diffractometric methods. These methods are quite useful and accurate for any defined feature such as acidity, organic/inorganic material content like alcohol or glucose content. However, main drawback of these methods are, they are mostly measurable for one feature, so the operational and maintenance costs are expensive. Image processing and Convolutional Neural Networks based methods are low-cost, but satisfactory results, however the main requirements of them are high computational power and this sometimes results overfitting problem. Moreover, some color tones are not detected correctly, hence they are detected and analyzed as crisp values. In this study, a hybrid methodology of computer vision and fuzzy inference systems has been proposed for a computationally effective, and accurate results for the determination of pH levels in liquid raw materials. The chemical sensor based color change has been detected by computer vision algorithms, then the required Red-Green-Blue (RGB) values have been transformed into Hue-Saturation-Value (HSV) space. These values are then input to Adaptive Network-based Fuzzy Inference Systems (ANFIS) to determine the pH level of any liquid material. During the first phase of the analyses, satisfactory results have been obtained and the research studies are ongoing within National Scientific and Technological Research Council of Turkey (TUBITAK) 1501-called R&D project.