Answers are grounded in this paper and cite their sources.
Demonstrates high accuracy in gas classification and concentration prediction, suggesting improved efficiency for industrial monitoring.
In industrial emission monitoring, traditional electronic nose systems treat gas classification and concentration prediction as separate tasks, which leads to error accumulation and high deployment costs. Existing temporal modeling methods mostly use single-scale feature extraction and cannot capture both global long-range dependencies and local detailed features at the same time. To address these challenges, we propose MT-TERN, a multi-task temporal feature learning framework for parallel joint modeling of gas species identification and concentration prediction. The framework introduces a ResNet1D-based residual temporal encoding module that extracts local temporal features using dilated 1D convolution and residual connections, without needing traditional multi-layer TCN cascades. We build a dual-layer feature extraction backbone network by connecting the self-attention temporal encoding layer and the residual temporal encoding layer in sequence, which extracts multi-scale temporal features of gas responses from both global and local views. By sharing the dual-layer encoding backbone and combining classification and regression tasks, both tasks are optimized together in a unified feature space, avoiding the error accumulation problem in serial architectures. Experimental results on two public datasets show that all classification accuracies exceed 0.99, with concentration regression R 2 above 0.980, while training efficiency improves by 52% compared to independent task models. This method offers an efficient and accurate solution for industrial gas detection with broad application potential in safety monitoring scenarios.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: