Assessment of waterborne metal contaminants by integrating deep learning with support vector machine and random Forest classifiers through Lemna minor phenotyping

Water is an important source of life for living beings on the earth. Due to urbanization, growth in industries and various other human activities, this water is polluted by microbes, pesticides, chemical solvents and metal contaminants. Such polluted water may degrade the quality of the crops and yield when utilized for irrigation. This research proposes a novel methodology to analyse the quality of water with respect to metal contaminants such as Arsenic (As), Copper (Cu), Lead (Pb) and Mercury (Hg) using Lemna Minor as a bio-indicator. Lemna minor was grown on known metal contaminated water samples as well as water collected from reservoir. The images of Lemna Minor grown on this medium were processed for the study. Image processing and machine learning techniques are integrated to assess the metal contaminants in water. In this research, VGG16 and VGG19 models are attempted initially. Due to over fitting issues encountered with these models, integrating the features of VGG 16 model with multiclass classifiers is attempted. VGG16 with Random Forest classifier produced an accuracy of 98.41 % whereas with VGG16 with SVM classifier produced an accuracy of 99.77 % in assessing the metal contaminants in water. The modified VGG- SVM model is extended for 12 class classification for assessing suitability of water for irrigation produced a classification accuracy of 84 %. Implementation of these hybrid methodology is first of its kind for the identification of metal contamination in water through image processing using the Lemna Minor phenotyping.