Water samples for turbidity detection
Computer vision and image processing have made significant progress in many real-world applications, including environmental monitoring and protection. Recent studies have shown that computer vision and image processing can be used to quantify water turbidity, a crucial physical parameter in water quality assessment. This paper presents a procedure to determine water turbidity using deep learning methods, specifically, convolutional neural network (CNN). At first, water samples were located inside a dark cabin before digital images of the samples were captured with a smartphone camera. A total of 71 samples were taken, representing varying magnitudes of nephelometry turbidity unit (NTU) between 0 to 100. In this research, CNN models were used to detect water turbidity. Based on the results, when grayscale images were used, and an NTU of less than 5 are considered images of clean water, the CNN algorithm achieved an accuracy of 0.9811 and a loss of 0.0514. When color images were used and an NTU less than 5 is considered clean water, the CNN algorithm achieved an accuracy of 0.9811 and a loss of 0.0414. The exact process was repeated when grayscale and color images were used, with NTU less than 15 considered as clean water. The performance of Python and R in terms of time required to train a convolutional neural network using Keras was compared. The results found that training a CNN Keras model in Python is faster than R, while the accuracy of the provided model is independent of the programming language
These are raw images of water samples with specific turbidity values prepared by mixing 1000 NTU Formazine stock solution with pure water.