A combined computer vision and convolution neural network approach to classify turbid water samples in accordance with national water quality standards

Computer vision (CV) technologies have been applied extensively in turbid water assessment. However, the accuracies of CV to detect turbidity is limited by several factors such as inferior image quality and adoptation of traditional machine learning (MsL). Several past studies have shown that the accuracy of CV to detect turbidity can be enhanced by employing image pre-processing and artificial intelligence (AI). Therefore, this study proposes a combined CV and convolution neural network (CNN) approach for turbid water classification in accordance with national water quality standards. A total of 71 turbid water samples in the range of 0–100 nephelometry turbidity units (NTU) were prepared from a mixture of 1000 NTU stock Formazine solution and distilled water. Digital images of water samples were acquired with the Asus Zenfone Go smartphone and nephelometry principle. Synthetic minority oversampling technique (SMOTE) was employed for image pre-processing, synthesis, and augmentation. Classification models were developed with Keras CNN architecture and TensorFlow framework. Two national water standards referred to were Malaysia National Water Quality Standard Class I (NWQS-C1) and France Système D’évaluation de la Qualité des cours d’EAU bleu aptitude (SEQ-EAU-b). The proposed CV-CNN approach was successfully implemented with 94.34–98.42% accuracy. The accuracy of CNN was slightly higher when trained with color images (98.42%) than with grayscale images (94.34%). This study demonstrated that CV and CNN are excellent tools for water quality assessment, especially for classifying water turbidity.

Text 115734 (2).pdf - Published Version Restricted to Repository staff only Download (1MB) Official URL or Download Paper: https://link.springer.com/article/10.1007/s13369-0...

Actions (login required)

View Item