Thermal Anomaly Detection using Physics Aware Neuromorphic Networks: Comparison between Raw and L1C Sentinel-2 Data
Authors: Stephen Smith, Cormac Purcell, Gabriele Meoni, Roberto Del Prete, Zdenka Kuncic
Organizations: School of Physics, University of Sydney, Sydney, Australia · BolgianTen Ltd., Liverpool, United Kingdom · School of Computer Science, University of New South Wales, Sydney, Australia · Φ-lab, European Space Research Institute (ESRIN), European Space Agency (ESA), Frascati, Italy · Advanced Concepts and Studies Office, European Space Research Institute (ESRIN), European Space Agency (ESA), Frascati, Italy · Emergentia Pty. Ltd., Sydney, Australia
Damage caused by bushfires and volcanic eruptions escalates rapidly when detection is delayed, making fast and reliable early warning capabilities essential. Recent Earth Observation (EO) approaches have shown that thermal anomaly detection can be performed directly on decompressed Level-0 (L0) sensor data, avoiding computationally expensive preprocessing chains. However, direct exploitation of raw data remains challenging due to domain shift, sensor drift, radiometric inconsistencies, and the scarcity of labelled training samples. To address these challenges, this work proposes a Physics-Aware Neuromorphic Network (PANN) framework for onboard thermal anomaly detection. The proposed lightweight architecture, inspired by physical neural network principles and neuromorphic computing paradigms, is evaluated using two Sentinel-2 datasets: decompressed L0 with additional metadata (i.e. raw) and Level-1C (L1C). The PANN achieves a Matthews Correlation Coefficient (MCC) of 0.809 on raw measurements, compared to 0.875 when using ground-processed L1C products. The mean processing latency per L0 granule is 2.44±0.09s, which is below the Sentinel-2 acquisition time of 3.6s, demonstrating the feasibility of real-time, onboard processing. Furthermore, the projected execution time for the corresponding neuromorphic hardware instantiation is substantially lower at 0.1290±0.0002s. Memory usage, including all necessary programs and packages, remains within realistic onboard constraints, with requirements of 0.673±0.007Gb for the software PANN and 0.393±0.004Gb for the estimated hardware realisation. Overall, these results indicate that PANN offers a promising pathway toward low-latency and resource-efficient onboard EO processing for thermal event detection.
Figures & tables
Figure 1: Visualisation of the simulated nanowire network device model used in PANN, showing the input electrodes (large blue circles), readout electrodes (large red circles) and the underlying nanowire network underneath with the nanowires (grey lines) and nanowire–nanowire junctions (small grey dots). Scale bar is 50μm .
Figure 2: Diagram of the model pipeline. Starting with the input image (left), the image is fist normalised and tiled into smaller 128×128 pixel patches, followed by a 16×16 max pooling layer. The two spectral bands, B8A and B13, are split as denoted by ⨂ and fed into identical PANNs. Output feature vectors are concatenated with a skip connection, denoted by ⨁ . Finally, the spanning normal distances between the two feature vectors is calculated and a threshold value is applied to classify each tile and produce a predicted output map (right).
Raw Data
L1C Data
MCC
0.809
0.875
Precision
0.763
0.887
Recall
0.867
0.869
Table 1: MCC, precision and recall results for the PANN model for both raw and L1C datasets.
Figure 3: Confusion matrix for the raw (a) and L1C (b) datasets, showing the number of samples and the percentage for that class. The background colour shading represents the class percentage, with white for 0% and black for 100%.
Figure 4: Effect of varying the threshold value on the different evaluation metrics; MCC (blue), precision (orange) and recall (green). Results for the raw data are shown with a solid line (x-axis values on bottom) and the L1C data results are shown with a dashed line (x-axis values on top).
Figure 5: Three examples for fires (a) and volcanoes (b) for both the raw and L1C datasets. Each example shows an RGB image, using bands B8A, B11 and B12 (top row), along with the target output (middle row), where tiles containing thermal anomalies are shown in red, and the predicted output (bottom row), where tiles predicted by the PANN to have thermal anomalies are shown in orange.
PANN model
Estimated Hardware
Time (s / granule)
2.44±0.09
0.1290±0.0002
Memory (GB)
0.673±0.007
0.393±0.004
Table 2: Compute resources calculated for the PANN model and estimated for the hardware nanowire network device, including time to process a granule in seconds and the total memory requirement of the model and all required software and packages. For all values the ± SEM is over 5 runs.
We present a deployed system for on-orbit wildfire detection aboard a nine-satellite commercial thermal infrared constellation, operating under demanding joint constraints: sub-megabyte model footprint, sub-150 ms per-batch TensorRT FP16 inference on an NVIDIA Jetson Xavier NX, and an end-to-end alert pipeline targeting under 10 minutes from satellite overpass to fire event communication. The system operates on uncalibrated mid-wave infrared (MWIR) single-band imagery at 200 m ground sampling distance, where fires frequently appear as sub-pixel or single-pixel thermal anomalies under extreme class imbalance -- challenges not addressed by the contextual thermal-thresholding pipelines (MODIS, VIIRS) that currently dominate operational fire monitoring. We present an empirical study of lightweight dense representation learning for this regime using a proprietary nine-satellite MWIR dataset. We compare dense masked autoencoding (DenseMAE) and a hybrid DenseMAE+EMA (exponential moving average) distillation variant, and evaluate representations via linear probing and full-distribution pixel-level average precision (AP) under extreme class imbalance. DenseMAE pretraining enables compact downstream models on the latency-accuracy Pareto frontier: our fastest SSL-pretrained model achieves 0.640 test AP and 0.69 event-level Fire-F1 with 65.34 ms latency per batch and a 0.52 MB engine, without pruning or compression. The best configuration reaches 0.699 AP and 0.744 Fire-F1 below 1 MB, outperforming a supervised baseline (0.650 AP) under comparable constraints.
Matthias Rötzer, Veronika Pörtge, Martin Ickerott +6
In this work, we present LIANet (Location Is All You Need Network), a coordinate-based neural representation that models multi-temporal spaceborne Earth observation (EO) data for a given region of interest as a continuous spatiotemporal neural field. Given only spatial and temporal coordinates, LIANet reconstructs the corresponding satellite imagery. Once pretrained, this neural representation can be adapted to various EO downstream tasks, such as semantic segmentation or pixel-wise regression, importantly, without requiring access to the original satellite data. LIANet intends to serve as a user-friendly alternative to Geospatial Foundation Models (GFMs) by eliminating the overhead of data access and preprocessing for end-users and enabling fine-tuning solely based on labels. We demonstrate the pretraining of LIANet across target areas of varying sizes and show that fine-tuning it for downstream tasks achieves competitive performance compared to training from scratch or using established GFMs. The source code and datasets are publicly available at https://github.com/mojganmadadi/LIANet/tree/v1.0.1.
Mojgan Madadikhaljan, Jonathan Prexl, Isabelle Wittmann +2
University of the Bundeswehr Munich · IBM Research – Europe · Columbia University +1
Methane is a major driver of near-term climate change, and rapidly identifying its emission sources is a critical climate intervention. Spaceborne hyperspectral imagery is the primary tool for this task, but the volume of data produced by each sensor makes ground-based detection impractical and necessitates onboard detection. Classical methods incur prohibitive computational cost on onboard hardware, while deep learning models are fast but fall short on detection quality. We propose FLAME, a physics-guided neural operator that builds the physics of methane absorption directly into its architecture. On the methane detection benchmark, FLAME achieves the highest detection accuracy among all evaluated methods, reduces the pixel-level false positive rate by nearly 3× over the strongest neural baseline, uses the fewest parameters among learned baselines, and runs within the latency budget of onboard satellite hardware.