A spatial-temporal deep learning method for heavy metal ion detection

pose severe threats to public health and aquatic ecosystems. Traditional electrochemical detection faces significant challenges including high sample consumption, manual intervention, and difficulties parsing overlapping signals. To address these issues we propose an intelligent metal ion detection platform integrating a microfluidic chip with a deep learning framework. The microfluidic system enables precise droplet manipulation for automated signal acquisition utilizing minimal sample volumes. To overcome traditional limitations in direct feature extraction we developed a multidimensional relational feature enhancement architecture. Specifically, we propose a multi-task framework prioritizing the integration of a Convolutional Neural Network and a Temporal Convolutional Network to form the primary C-TCN model. This effectively models sequence dependencies to simultaneously achieve accurate ion classification and concentration regression. Experimental results show the optimal C-TCN model increased classification accuracy by 25.9% compared to the unmodified CNN, successfully outperforming alternative LSTM variants. Furthermore, the C-TCN framework significantly outperforms traditional machine learning methods across all critical evaluation metrics including F1 score, AUC, and regression error. These results demonstrate that the synergistic combination of spatial purification and temporal convolutional tracking provides a highly stable and robust paradigm for intelligent electrochemical sensing.