Source-linked AI summary

LDAC-Net: A Learnable Multi-Lag Differencing Attention-Convolution Network for Drift-Robust Recognition with Low-Cost MOX Gas Sensors

Xin Zhang, Liangxiu Han, Yue Shi, Tam Sobeih

arXiv:2608.25646v1cs.LGcs.AI

TL;DR

Low-cost MOX gas-sensor recognition must separate class-relevant dynamics from drift, scale changes, offsets, and cross-channel variation, while fixed differencing uses a manually chosen lag and can discard absolute response information. LDAC-Net learns window-conditioned normalisation and multi-lag differencing within an end-to-end attention-convolution model. It improves recognition across SmellNet-Base, SmellNet-Mixtures, and eNose-Drift comparisons, while remaining compact for modest datasets.

  • Problem

    MOX recognition is challenged by slow transients, drift, scale variation, offsets, and cross-channel correlations, while fixed FOTD uses a hand-picked lag and discards absolute response information.

  • Method

    LDAC-Net uses an LDFE front end with window-conditioned affine normalisation and learnable multi-lag differencing, followed by a compact attention-convolution backbone.

  • Results

    LDAC-Net outperforms comparison models across SmellNet-Base, SmellNet-Mixtures, and eNose-Drift, including 68.2 % Top-1 on SmellNet-Base.

  • Takeaways & Limitations

    Learnable, sensor-aware preprocessing can integrate drift compensation and dynamics extraction into compact MOX recognition models across datasets and sensor arrays.

  • Takeaways & Limitations

    Both evaluated corpora are modest in size, so larger and more varied datasets are needed to confirm that the reported gains hold.

Abstract

from arXiv · show

Portable electronic-nose systems based on low-cost metal-oxide (MOX) gas sensors offer a practical solution for gas and odour recognition, but their signals are affected by slow chemical transients, drifting sensor offsets, scale variation, and cross-channel correlations. Existing pipelines commonly use fixed first-order temporal differencing (FOTD), which requires a manually selected lag and may discard useful response information. We propose LDAC-Net, an end-to-end learnable multi-lag differencing attention-convolution network that operates directly on multi-channel MOX signals. Its learnable differential feature enhancement front-end combines window-conditioned statistical affine normalisation, which compensates for window-specific offset and scale variation, with learnable multi-lag differencing, which weights and combines temporal differences across multiple lags. A compact attention-convolution backbone subsequently models local transients and longer-range temporal dependencies. On the 50-class SmellNet-Base task, LDAC-Net achieves 68.2% top-1 accuracy, exceeding the best FOTD-preprocessed comparison model by approximately 14 percentage points and the raw-input Transformer by more than 30 points. Ablation studies confirm the contributions of both proposed components. The representation also transfers to SmellNet-Mixtures, improving accuracy from 45.4% to 50.5%, and generalises to the 62-channel eNose-Drift benchmark under strong long-term drift, achieving 70.6% top-1 accuracy and 69.6% macro-F1. These results outperform the best comparison model with dataset-retuned FOTD preprocessing by 8.0 and 3.0 points, respectively, demonstrating that learnable, sensor-aware preprocessing is more effective than fixed handcrafted differencing for low-cost MOX gas-sensor recognition.

1. Introduction

LDAC-Net addresses drift-robust recognition from low-cost MOX sensor windows by learning compensation and dynamics extraction inside an end-to-end model. Its front end combines window-conditioned affine normalisation with learnable multi-lag differencing, followed by an attention-convolution backbone.

  • Low-cost MOX sensors are practical but difficult to use because their signals combine slow transients, session drift, scale variation, and cross-channel correlations.
  • Fixed first-order temporal differencing suppresses slow drift and exposes dynamics but requires a hand-picked lag and may discard absolute response information.
  • LDAC-Net learns drift compensation, dynamics extraction, and classification jointly from raw multi-channel windows without external preprocessing or recalibration.
  • The compact attention-convolution backbone is designed to capture both local transients and longer-range temporal dependencies in sensor windows.
  • LDFE combines window-conditioned affine corrections from per-window statistics with learnable multi-lag differences, channel mixing, weighting, and gating.
  • A longer analysis window provides sufficient temporal context for in-model differences to expose response dynamics rather than noise.

2. Related Work

Related work addresses MOX recognition through handcrafted response descriptors, fixed drift preprocessing, and deep multivariate time-series models. LDAC-Net targets the remaining gap by learning window-conditioned correction and multi-lag differencing jointly inside the recognition model.

  • Earlier electronic-nose systems commonly extracted response-curve descriptors for shallow classifiers, while later approaches applied fixed preprocessing before classification.
  • FOTD suppresses slow drift and exposes transient dynamics, but its fixed lag applies uniformly across substances, channels, and sessions while discarding absolute signal level.
  • Deep-learning alternatives for multivariate time-series recognition include convolutional, recurrent, and attention-based models that capture temporal patterns in different ways.
  • Two gaps remain: preprocessing is often external and fixed, while non-stationarity methods may normalise statistics without learning multi-scale differencing for slow MOX dynamics.
  • LDAC-Net combines window-conditioned affine correction with learnable multi-lag differencing, unlike methods using only statistical rescaling or predefined difference kernels.

3. Method

LDAC-Net classifies raw multi-channel windows through a learnable front end that normalises window statistics and extracts dynamics across multiple lags. Attention-convolution blocks then combine local and long-range temporal modelling before attention-based readout.

  • Overview: LDAC-Net predicts substance labels directly from raw fixed-length multi-channel windows without external drift correction or handcrafted temporal differencing.
  • Learnable Differential Feature Enhancement: LDFE combines WSAN, which learns per-window affine corrections from channel statistics, with LMLD, which learns weighted multi-lag differences and channel interactions.
  • Attention-Convolution Backbone: The AC backbone pairs depthwise convolution for local structure with self-attention for longer-range context, while attention pooling weights informative timesteps during readout.
  • End-to-End Architecture: The LDFE output is projected into a T × d token sequence, processed by three AC blocks, and mapped to 50-way logits.
  • Window-conditioned Statistical Affine Normalisation: WSAN starts as instance normalisation, then learns residual affine corrections that remove drift while retaining a route for discriminative absolute response levels.
  • Learnable Multi-Lag Differencing: LMLD computes differences at lags 5, 10, 25, and 50, concatenates them with the identity stream, and applies learned mixing, normalisation, and channel gating.

4. Experiments

The evaluation tests LDAC-Net on primary substance recognition, ablations, raw-input model comparisons, and generalisation to mixtures and a distinct drifting sensor array. The experiments therefore assess both component contributions and transfer across tasks and hardware.

  • Dataset: SmellNet-Base is the primary 50-way substance-recognition benchmark using six-channel low-cost MOX recordings.
  • Dataset: SmellNet-Mixtures evaluates compositional ratio prediction over twelve odorants using four sensor channels, while eNose-Drift tests cross-array recognition under strong long-term drift.
  • Protocol: The eNose-Drift analysis uses an initial calibration, chronological training-to-testing splits, and Acc@1 plus macro-F1 reporting across repeated seeds.
  • Protocol: Raw-input configurations receive no external drift correction, smoothing, or handcrafted temporal differencing unless explicitly stated otherwise.
  • Experimental Design: The study evaluates primary performance against mainstream temporal baselines, component contributions through leave-one-out ablations, and comparisons with popular multivariate time-series models.
  • Experimental Design: Generalisation experiments test both transfer to a different task on the same array and transfer to an independent 62-channel MOX array with a different drift regime.

5.1. Model Performance

LDAC-Net performs best on SmellNet-Base using raw windows, surpassing FOTD-preprocessed baselines. Performance depends strongly on analysis-window length, peaking at T = 300.

  • 68.2% Top-1, 89.4% Top-5, and 65.2 macro-F1 make LDAC-Net the best model on every reported metric.
  • 14.4 points on Top-1 and 12.8 on F1 separate LDAC-Net from the strongest FOTD-preprocessed baseline despite no external preprocessing.
  • 57.8% at T = 100 rises to 68.2% at T = 300, then falls to 64.8% at T = 400.
  • At T = 300, LMLD contributes +5.9 points over LDAC-Net without LMLD, compared with −0.3 points at T = 100.
  • Longer windows provide sufficient context for slow MOX transients, while T = 400 yields too few windows per session for training.

5.2. Ablation Study

Ablations show that both window-conditioned normalisation and learnable multi-lag differencing materially contribute to LDAC-Net’s performance.

  • 14.0 points are lost when WSAN is replaced by plain InstanceNorm, falling from 68.2% to 54.2% Top-1.
  • 5.8 points are lost when the LMLD differencing front is removed, reducing Top-1 accuracy to 62.3%.
  • 6.2 points are lost by removing squeeze-excitation, while removing learnable channel mixing costs 4.9 points.
  • 4.6 points are lost when the multi-lag bank is collapsed to the single lag {25}, confirming the value of multiple lags.

5.3. Model Comparison

LDAC-Net outperforms the Transformer and published time-series alternatives on SmellNet-Base, including models using drift-oriented or temporal preprocessing strategies.

  • 52.61% is the strongest competing accuracy, achieved by the Non-stationary Transformer, which still trails LDAC-Net by more than 15 points.
  • 68.2% Top-1 makes LDAC-Net more than 14 points better than Transformer-FOTD and over 15 points better than the best raw-input alternative.
  • 50.28% for Autoformer and 50.47% for Neural-ODE indicate that their added capacity does not match LDAC-Net on this small training set.
  • 43.63% makes the dilated TCN the weakest published-method transplant, while the raw Transformer reaches 37.29%.

5.4. Generalisation

LDAC-Net generalises beyond 50-way recognition, transferring to compositional mixture prediction and a different, strongly drifting 62-channel sensor array.

  • Compositional Generalisation: 50.5% Top-1@0.1 exceeds the Transformer’s 45.4% on SmellNet-Mixtures, a 5.1-point gain.LDAC-Net also achieves the lowest MAE (0.050) and highest Top-K presence score (81.1%).
  • Cross-Dataset Generalisation: 70.6 ± 4.8 Acc@1 and 69.6 ± 3.9 macro-F1 are achieved on the 62-channel eNose-Drift benchmark.The benchmark differs in hardware and channel count and exhibits strong long-term drift over twelve months.
  • Cross-Dataset Generalisation: LDAC-Net exceeds the strongest dataset-retuned FOTD baseline by 8.0 Acc@1 points and 3.0 macro-F1 points.The model operates end-to-end on raw signals after scaling the analysis window and lag-bank range to the slower transient.

6. Discussion

The discussion identifies why adaptive differencing and window length matter, while noting data scarcity, a manually selected global window, and untested deployment conditions.

  • 6.1. Window-Conditioned Affine versus Fixed Differencing: Fixed differencing can under- or over-compensate drift and discard discriminative absolute response levels.WSAN preserves such information, while LMLD replaces FOTD’s fixed lag with a learnable multi-lag bank.
  • 6.1. Window-Conditioned Affine versus Fixed Differencing: 63.6% is the weakest lag-bank choice, and multi-lag benefits require coverage across a wide range of time scales.Banks restricted to only long or short lags are less effective than the broader bank used by LDAC-Net.
  • 6.2. Window Length as a Key Accuracy Factor: Increasing T from 100 to 300 raises Top-1 accuracy from 57.8% to 68.2%.Longer windows provide more temporal context for in-model differencing, while overlap can act as augmentation and regularisation.
  • 6.2. Window Length as a Key Accuracy Factor: LMLD’s benefit grows from +2.3 points at T = 200 to +5.9 at T = 300 and +7.2 at T = 400.The accuracy curve is unimodal because longer windows eventually reduce the number of windows; T = 300 is the peak.
  • 6.3. Limitations: Both evaluated corpora remain modest in size, so larger and more varied datasets are needed to confirm that the gains hold.SmellNet-Base has about 2,650 training windows at T = 100 and about 1,650 at T = 300.
  • 6.3. Limitations: The global T = 300 window is fixed by offline grid search even though the best length may vary by substance, channel, and response speed.The paper proposes input-adaptive or multi-scale windows as a next step.
  • 6.3. Limitations: Transfer has been tested only under controlled offline conditions, and on-device latency and power remain unmeasured.This includes the extra cost of the per-window WSAN pass.

7. Conclusion

LDAC-Net integrates drift removal and dynamics extraction into a compact end-to-end model, achieving strong recognition and transfer across tasks and sensor arrays.

  • Architecture: LDAC-Net classifies raw low-cost MOX sensor windows end to end using learnable drift compensation and temporal dynamics extraction.Its LDFE front end combines WSAN and LMLD, while the AC backbone uses convolution and self-attention.
  • Main Results: 68.2% Top-1 is achieved on the 50-way SmellNet-Base task, with 89.4% Top-5 and 65.2 macro-F1.Top-1 is more than 14 points above the strongest FOTD-preprocessed baseline and over 30 points above the raw-input Transformer.
  • Main Results: Increasing T from 100 to 300 improves Top-1 accuracy by 10.4 points, from 57.8% to 68.2%.The ablation study reports that each evaluated component contributes to classification accuracy.
  • Generalisation: Generalisation experiments show transfer across tasks and sensor arrays.The conclusion identifies SmellNet-Mixtures and eNose-Drift as the two transfer settings.
Loading 2608.25646v1…