Detection of surface water pollution with full absorption spectroscopy: Integrating 2D image conversion and ResNet...

Global industrialization and urbanization have caused significant degradation of water environments, posing major risks to human health, ecological stability, and sustainable economic development. Traditional water quality monitoring methods are becoming insufficient to address these complex challenges. This study presents a novel MGR (MSC-GAF-ResNet) model with full-spectrum absorption data for surface water pollution classification and monitoring. The methodology employs Savitzky-Golay filtering and Multiplicative Scatter Correction for preprocessing, converts spectral data into two-dimensional representations using Gramian Angular Field techniques, and applies a residual convolutional neural network for classification. Experiments on 301 water samples from the Jiulong River Basin demonstrated that the MGR model achieved 97.7% classification accuracy with high computational efficiency, outperforming conventional machine learning methods (SVM and LSTM). Integration of temporal and geographical metadata enables effective pollution source tracing, improving environmental monitoring and emergency response capabilities. This research provides an efficient approach to water quality monitoring and pollution source identification, supporting sustainable water resource management.