Statistical Estimation of Lead (Pb) in Groundwater from Smartphone Images of Paper Test Strips using CIEDE2000 Model

Objectives: The propose research work explores an image based data analysis and interpretation approach for detection and estimation of Lead (Pb) in ground water to minimize the human error. It also reviewed the existing traditional techniques which involve chemical analysis or test strips in laboratory. Most of them are expensive, time-consuming and based on human perception. They detect only the presence of the contaminated heavy metals but fail to estimate the amount. Methods: This proposed work utilizes images of paper based test strips that are captured by a smartphone camera under ambient lighting condition. The RGB values of the captured images are converted to CIEDELAB color space. Image features were extracted and the perceptual color difference (ΔE) was calculated. Various statistical regression techniques are used to find the correlation between color metrics and the Lead (Pb) concentration. Findings: The study suggests that the smartphone-based images can serve as a cost-effective means for on-site heavy-metal screening. Comparative analysis of linear, logarithmic, exponential and quadratic trendlines suggests a consistent relationship between the analyst (Pb) amount and CIEDE2000 value. Based on the improved coefficient of determination (R2) value, Linear and Quadratic trendline indicates the most supportive relationship between lead (Pb) quantity and CIEDE2000 values. The outcome established the system as a practical alternative under non-uniform lighting conditions in comparison to the conventional RGB-based methods reported in the literature. This approach neither requires any sophisticated instruments or any complex programming algorithms for data analysis. Novelty: The study is based on CIEDE2000 perceptual color difference metric unlike relying completely on absolute RGB values or controlled imaging arrangements as done in most of the reported methods in literature. This approach reduces data dependency and improves practical suitability for on field monitoring in resource constrained situation. Keywords: Feature extraction, Statistical modeling, Regression, Heavy metals, Image Processing