{"@context":"https://schema.org","@type":"ScholarlyArticle","headline":"Machine learning models for measurement of pH using a low-cost image analysis strategy","author":[{"@type":"Person","name":"B. X. Ferreira"},{"@type":"Person","name":"Alline V. B. de Oliveira"},{"@type":"Person","name":"João Cajaiba"},{"@type":"Person","name":"Vinicius Kartnaller"},{"@type":"Person","name":"Brunno Ferreira dos Santos"}],"datePublished":"2025","isPartOf":{"@type":"Periodical","name":"Measurement Science and Technology"},"identifier":"https://doi.org/10.1088/1361-6501/adffa0","description":"Abstract\n Soft sensor use has been growing in many industries as an alternative to conventional physical sensors due to their disadvantages, such as cost and maintenance. One difficult measurement is the pH values in a pressurized system, which requires specialized equipment. This work aims to develop machine learning models to directly determine the pH in pressurized systems (up to 6 MPa) from the reaction medium using image analysis as an initial step, creating an applicable soft sensor. Classification and prediction models were developed using different techniques: convolution neural networks (CNNs), support vector machines (SVMs), and decision trees (DTs). All of them were explored in classification models, but CNN was the only one used for the regression models. The models were trained with buffer solutions images and then tested using three scenario cases (acid–base titration curve, pressurized CO2–H2O systems, and brightness alteration). The best classification models were the CNN ones, as the SVM and DT had good results for the test group (accuracy (ACC) above 90%), but the mean ACC values were below 75% for the scenario cases. Model CNN_class_10 presented the best results (accuracy above 70% in all cases). The CNN prediction models showed the best results, with the CNN_pred_7 model achieving R\n 2 values higher than 80% for all experiments in the first two scenario cases and exhibiting excellent stability to augmentation variations (brightness and rotation). In conclusion, improving the CNN regression models appears to be the most promising approaches to creating soft sensors in future works.","url":"https://innscience.ca/en/community/literature/doi_10_1088_1361_6501_adffa0"}