Paper Title
Development of Image Recognition Technology for Wastewater Turbidity

Abstract
Clean water resources are indispensable to humanity. Traditional turbidity measurement methods used in water treatment plants are time-consuming, rely on expensive instruments, and produce data formats that are not compatible with other devices. This study aims to develop an alternative approach to traditional turbidity measurement methods. Specifically, it leverages deep learning with a convolutional neural network (CNN) model for image recognition of water samples with varying turbidity levels. A regression task is performed on continuous turbidity value data to estimate water quality turbidity. The results of the turbidity recognition model indicate a high correlation between estimated and actual values, with an R² of 0.988 and a P-value of 0.009, demonstrating statistical significance. The maximum turbidity value captured was effectively increased by reducing the container's inner diameter length, which enhanced the range of turbidity samples used for model training. Heatmaps revealed that when turbidity values reached approximately 180 NTU, the CNN was unable to estimate turbidity based on image features. To assist water treatment plants in using image recognition for water quality monitoring, this study developed an image-based turbidity recognition technology. This approach aims to provide a convenient, fast, and low-cost alternative to traditional measurement methods. Keywords - Turbidity, Deep Learning, Convolutional Neural Network, Image Recognition