Machine learning approach for measuring water quality of coastline and estuaries in Chennai coastal area

Abstract

Chennai has had the longest coastline over other major cities of India. It is decidedly vital to monitor seawater quality due to the increased coastline population. This study presents an Android mobile application based on a machine learning approach to perform basic testing parameters of seawater by applying convolutional neural network concepts. A commercially available “saltwater master test kit” was used in this study to test the level of pH, Ammonia, Nitrite, Nitrate, and Total Phosphate (T.P.) in seawater. Six water samples collected from every 10 regions, including the coastline and coastal estuaries, were tested with the test kit in a microplate. Images were captured in the solely designed mobile app, and they were pre‐processed, and RGB (Red, Green, and Blue) were recorded from the Region of Interest (R.O.I.) of the image. A supervised Convolutional Neural Network Image Classifier (SCNNIC) algorithm was developed to classify the RGB pixel values. CIEDE2000 (C2K) color difference algorithm was applied over the recorded RGB values with the datasets stored previously to result in the nearest color match between the ideal dataset and R.O.I. of the captured image. Grayscale and RGB methods results were compared with the standard APHA method. This C2K color difference algorithm produced a percent accuracy of over 98% compared with other methods used, and R2 recorded by curve fitting method for pH, T.N. and T.P. were above 0.98. Disquieting results were reported in this study, especially in Muttukkadu backwater and Adayar river backwater estuaries, reported high values of pH (7.82 and 8.17), TN (13.74 mg/L and 13.45 mg/L) and T.P. (0.266 μg/L and 0.724 μg/L). The mean for all 10 regions of 110 km chosen in this study for the 2 years was calculated, and values were obtained as pH‐8.33, TN‐5.321 mg/L, and TP‐0.143 μg/L.