Labeled RGB Image Dataset for Water Turbidity Classification Based on Total Suspended Solids Using Deep Learning

This dataset contains labeled RGB images of water samples with varying turbidity levels, generated under controlled laboratory conditions using suspended clay particles as the primary source of turbidity. The samples were prepared from 33 laboratory-generated mixtures with controlled concentrations of total suspended solids (TSS), following a turbidity classification scheme inspired by environmental standards. The dataset includes five turbidity classes, ranging from excellent water quality to highly contaminated conditions. Image acquisition was performed using a standardized experimental setup, including controlled illumination, fixed camera positioning, and continuous agitation to maintain particle suspension. Turbidity levels were validated using calibrated measurement instrumentation. Images were extracted from video recordings to capture temporal variations in turbidity patterns. The dataset was used to train and evaluate deep learning models based on convolutional neural networks, particularly EfficientNet-B0 with transfer learning, achieving high classification performance. This dataset supports research in computer vision and artificial intelligence for water quality assessment, providing a scalable and low-cost alternative for environmental monitoring applications.