Can Infrared Thermography Read Gunshot Residue Patterns?

A forensic measurement problem: can a transient thermal response reveal information about a residue pattern without destroying the evidence?

A textile may hold evidence an examiner wants to study without altering it. Chromogenic colour tests consume the residue they reveal. Tape-lifts for SEM-EDS remove particles permanently. The question researchers have begun asking is whether the fabric can reveal information about a gunshot event through physics — heat, rather than chemistry — before any of that happens.

A shirt arrives at a laboratory with a suspected bullet hole. Before it is treated with a reagent, dusted, or sampled for scanning electron microscopy, an examiner faces a quiet problem: nearly every standard next step changes the evidence in some way. If a case later needs a second look, the fabric may no longer hold the same answers.

This is the motivation behind a paper published in Forensic Science International in July 2026. A team led by researchers at the Faculty of Biomedical Engineering, Czech Technical University in Prague — working with ballistic data generated at the New Technologies Research Centre (NTC) of the University of West Bohemia in Pilsen — asked whether a brief flash of light, an infrared camera, and a few seconds of cooling could extract measurable information about a firearm discharge from a textile target, without sampling or chemically treating it at all [1].

The honest answer is: sometimes, under specific and clearly bounded laboratory conditions the authors themselves are careful to state. This article explains what flash-pulse infrared thermography (IRT) actually measures, what the July 2026 study found, and — just as importantly — what it did not establish.

1. The Forensic Problem: Evidence You Don't Want to Destroy

Gunshot residue (GSR) evidence on clothing is fragile in two senses. Physically, particles sit loosely on textile fibres and can be dislodged by folding, friction, or airflow. Evidentially, most confirmatory methods are consumptive by design: adhesive-stub lifts for SEM-EDS remove the particles under study; the Modified Griess Test and Sodium Rhodizonate Test chemically react with residue to produce a visible stain, permanently altering the sample [2]. SEM-EDS remains the accepted standard for confirming inorganic GSR because it can resolve both particle morphology and elemental composition — the characteristic lead–barium–antimony (Pb-Ba-Sb) signature of conventional primers [3]. But it is slow, requires vacuum-chamber instrumentation, and cannot be repeated indefinitely on the same sample [1].

Traditional shooting-distance estimation compounds the problem: it often depends on visual assessment of powder tattooing and soot spread, an assessment that is subjective and breaks down on dark, patterned, or blood-contaminated fabric where residue is simply hard to see [4][5]. The growing use of lead-free, heavy-metal-free primers further erodes the classic elemental fingerprint that SEM-EDS depends on [6]. A rapid, non-destructive, objective screening step performed before any consumptive method — one that preserves the sample for SEM-EDS afterward — is therefore an attractive idea, provided it does not overstate what it can do.

2. What Is Gunshot Residue Actually Made Of?

GSR is not one substance; it is the physical and chemical debris of a small, controlled explosion. When a primer is struck, it detonates a charge — typically lead styphnate, barium nitrate, and antimony trisulfide in conventional Boxer-type primers — which ignites the propellant, largely nitrocellulose-based smokeless powder [1][7]. The resulting discharge produces several distinct categories of residue:

• Inorganic GSR (IGSR) — microscopic, often sub-3 µm spherical or irregular particles formed by vaporisation and rapid re-solidification of primer metals; the Pb-Ba-Sb particles central to SEM-EDS confirmation [7][8].
• Organic GSR (OGSR) — nitrocellulose, nitroglycerin, and stabiliser compounds from the propellant, typically characterised by chromatographic or spectroscopic rather than SEM-EDS methods [3].
• Soot and partially or fully burnt powder particles, which physically deposit as visible blackening and stippling around the entry site.
• Metallic wipe transferred from the projectile jacket and firearm barrel.
• Background particulates — brake-lining dust and certain industrial residues are known to mimic Pb-Ba-Sb signatures, which is one reason morphology and case context still matter alongside elemental data [7].

Non-toxic, heavy-metal-free ammunition (replacing lead, barium, and antimony with elements such as zinc, titanium, and strontium) is actively eroding the classic elemental fingerprint, which is part of the motivation for exploring detection principles that do not depend on that specific chemistry [1][6].

3. The Unexpected Idea: Read GSR Through Heat

The idea is not that GSR emits detectable heat on its own. It is that GSR changes how a textile responds to an external heat pulse. The basic sequence is:

flash → thermal response → cooling behaviour → infrared detector → thermal map → forensic interpretation

A short, intense pulse of light heats the fabric surface almost instantaneously. Every point then cools at a rate governed by its local thermal properties — how well it conducts heat away, how much heat it stores, and how efficiently it radiates energy as infrared light. Where GSR particles, soot, and fused residue sit on the fabric, those local properties differ from the surrounding clean textile. An infrared camera watching the surface cool can register that difference as a spatial thermal-contrast pattern, even where the residue is too faint or too obscured by dark colour or staining to see with the naked eye [1][9].

This is distinct from passive thermography, which simply observes a scene's existing heat with no imposed excitation. Active thermography deliberately injects thermal energy and studies the transient response. Flash-pulse thermography is a specific, well-established branch of active thermography — long used in industrial non-destructive testing to find subsurface defects in composite materials — repurposed here for GSR detection [9][10]. It is not a new invention; it is an established industrial inspection technique applied to a new type of "defect": residue deposition.

4. How Flash-Pulse Infrared Thermography Works

1. Thermal excitation — a flash lamp delivers a brief, high-energy light pulse to the fabric surface.
2. Infrared imaging — a cooled infrared camera records surface temperature at high speed immediately after excitation.
3. Transient response — each pixel's temperature rises and then decays; the shape of that decay curve encodes local thermal properties.
4. Thermal contrast — residue-affected regions cool differently from clean fabric, producing measurable contrast.
5. Image formation — sequences of frames are combined (through frequency-domain or statistical transforms) into single representative images.
6. Pattern extraction — the resulting image is analysed for spatial structure: the size, shape, and intensity distribution of the thermal anomaly.

5. What the July 2026 Study Actually Tested

The study, titled "Preliminary evaluation of gunshot residue pattern analysis using flash-pulse infrared thermography and multi-task deep learning," was published as an open-access article in Forensic Science International, Volume 384 (July 2026), article 112938, with an online publication date of 26 March 2026 [1]. The data-acquisition work was carried out under the MEDEPOZ project ("Methodology for rapid contactless and non-destructive detection of gunshot residues," OPSEC VK01010037), funded by the Ministry of the Interior of the Czech Republic and led by the University of West Bohemia, with partners including the Institute of Criminalistics and the Faculty of Biomedical Engineering at the Czech Technical University in Prague [1].

Parameter	Reported detail
Authors	Marek Sokol, Jan Hejda, Petr Volf, Martin Valenta, Rudolf Vávra, Patrik Kutílek [1]
Lead affiliation	Faculty of Biomedical Engineering, Czech Technical University in Prague [1]
Journal / DOI	Forensic Science International, 384:112938 · 10.1016/j.forsciint.2026.112938
Peer review / access	Peer-reviewed, open access (CC BY-NC-ND)
Data acquisition site & period	New Technologies Research Centre (NTC), University of West Bohemia, Pilsen; June 2023 – October 2025 [1]
Textile substrate	Single substrate only: white pre-shrunk bio-cotton, 380–390 g/m², ~12% dark yarn flecks, 250×250 mm targets [1]
Ammunition	9×19 mm Parabellum, Sellier & Bellot; FMJ (n=162) and Hollow Soft/Point (n=125) [1]
Firearms	5 models retained after filtering: Glock 43X, CZ 75D Compact, HK SFP9 (pistols); CZ Scorpion EVO3 S1, HK MP5 (SMGs) [1]
Shooting distances	5–100 cm, perpendicular (90°) only; mean 37.0 cm, SD 18.9 cm [1]
Sample size	392 collected; 312 retained after excluding rare classes and unknown-ammunition samples [1]
Thermal camera	FLIR A6751, 640×512 px, NETD ≤ 20 mK, 100 Hz for 15 s (1,500 frames/sample) [1]
Excitation source	Hensel EH Pro 6000 flash lamp, 360 mm standoff [1]
Chemical confirmation	Not performed on these samples (no independent SEM-EDX/ICP-MS); ground truth relied on known firing conditions [1]

6. Why Would Gunshot Residue Produce a Thermal Signature?

The July 2026 paper does not independently establish the underlying mechanism; it explicitly relies on and cites the foundational thermographic protocol work by Moskovchenko, Švantner, and Honner, describing two proposed physical explanations [1][9]:

• Emissivity change (proposed explanation, drawn from prior work): the blackening caused by soot and GSR deposition increases the local surface emissivity relative to clean fabric, so the affected area absorbs more of the flash energy and heats more than the surrounding textile.
• Physical fusion (proposed explanation): some residue elements solidify and fuse with textile fibres, forming a thin coating whose thermal conductivity, heat capacity, and diffusivity differ from unaffected fabric, altering the local cooling rate.

Both explanations are physically plausible and consistent with established principles of thermal contrast in active thermography, but neither was independently verified against chemically confirmed GSR in the July 2026 dataset — the authors state explicitly that no SEM-EDX or ICP-MS confirmation was performed on the thermographic samples used, so the model may in part be responding to fabric damage, mechanical deformation, or ordinary soot rather than chemically specific residue [1]. This is our interpretation of a stated limitation, not an additional finding: the paper is careful to flag it as an open question for future work.

7. Can Thermal Images Reveal Firing Distance?

Shooting distance was treated as a continuous regression target (5–100 cm) and, separately, as four categorical bins: CONTACT (≤20 cm, n=90), CLOSE (21–40 cm, n=133), MEDIUM (41–60 cm, n=96), and FAR (>60 cm, n=23) [1].

Approach	Best result (5-fold cross-validation)
Best traditional-ML baseline (distance bin)	73.9–75.6% accuracy [1]
Best single-task regression (feature-based, top-25 features, Random Forest)	MAE = 6.07 cm, R² = 0.778 [1]
Final multi-task hybrid model (regression, alongside 3 classification tasks)	MAE = 7.92 cm, R² = 0.653 [1]

These figures describe performance on a held-out fold of the same controlled experimental dataset — same textile, same ammunition manufacturer and caliber, same perpendicular firing angle. Radial-spread features (how far thermal contrast extends from the wound centre) were the most informative signal for distance, consistent with the well-established physical relationship between residue spread and range [1][11]. Laboratory classification of this kind is not the same as casework determination: an unknown crime-scene sample would need a comparably built reference dataset with matching firearm, ammunition, and substrate before this specific model's output could be treated as informative distance evidence.

8. Can It Distinguish Weapon or Ammunition Characteristics?

The final hybrid multi-task model achieved 94.3% (±1.1%) accuracy for binary weapon category (pistol vs. submachine gun), 85.1% (±3.7%) for identifying which of five specific firearm models fired the shot, and 92.6% (±4.7%) for distinguishing FMJ from Hollow Soft/Point ammunition [1]. These are classification results within a closed, known set of five firearms and two ammunition types the model was trained on — not individual identification of an arbitrary, previously unseen firearm from an unknown crime-scene sample. The distinction matters: a classifier that can correctly sort five specific, previously characterised firearms cannot, without further validation, be assumed to generalise to a sixth firearm it has never seen, or to a different textile, ammunition brand, or shooting angle [1].

9. IRT vs SEM-EDS: Complement or Competitor?

Feature	Flash-Pulse IRT	SEM-EDS
Destructive sampling	No — non-contact, non-destructive [1]	Yes — tape-lift removes particles [3]
Chemical composition	Not measured	Directly measured (elemental) [3]
Spatial pattern	Captured as thermal contrast map [1]	Captured at particle level, smaller field
Particle morphology	Not resolved at particle scale	Resolved (µm-scale imaging) [3]
Speed	~15 seconds acquisition [1]	Hours, specialised lab required [3]
Screening role	Proposed as preliminary screening [1]	Confirmatory / evidentiary standard
Confirmatory analysis	Not established as confirmatory	Yes — accepted gold standard [3]
Operational maturity	Proof-of-concept / laboratory stage [1]	Routine, standardised (ASTM E1588) [3]
Validation	Single-lab, single-dataset so far [1]	Decades of inter-laboratory validation

The authors are explicit on this point: the technique "should be understood as a complementary screening tool that could precede or supplement, but not replace, established chemical confirmation methods such as SEM-EDX" [1].

10. What Could This Change at a Crime Scene?

If validated further, the realistic near-term applications are modest and specific: rapid triage of multiple textile items to prioritise which should go to SEM-EDS first; documentation of a residue pattern before any consumptive step is applied; and visualisation of patterns on dark fabric where visual and colour-test methods struggle [1][4]. None of these applications require IRT to identify a firearm or determine a distance on its own — they position it as a fast, non-destructive first look, not a final answer.

11. The Problems Researchers Still Have to Solve

The authors' own limitations section is unusually candid, and includes:

• A single textile substrate — white bio-cotton only; no dark, patterned, synthetic, or blended fabrics were tested, despite dark-fabric visualisation being one of the technique's main claimed advantages over colour tests [1].
• A single ammunition manufacturer and caliber (Sellier & Bellot 9×19 mm); results may not generalise to other primer chemistries, especially lead-free formulations [1].
• Perpendicular (90°) shots only — no oblique angles, which are common in real casework.
• A small, imbalanced dataset (312 samples across 5 firearm models), with one model (Glock 43X) making up 52% of the data.
• No independent chemical confirmation (SEM-EDX or ICP-MS) of GSR on the thermographic samples themselves.
• No inter-laboratory, cross-camera, or cross-device reproducibility testing.
• No environmental variation — temperature, humidity, ageing, or contamination effects were not evaluated.
• No testing on casework-derived samples; data was collected entirely under controlled laboratory conditions.
• Interpretability analysis (GradCAM, Integrated Gradients) was qualitative only, with no quantitative faithfulness testing [1].

The authors themselves conclude the underrepresentation of far-distance samples (n=23 beyond 60 cm) likely degraded performance at greater range, and that the model may be exploiting setup-specific artefacts (camera noise, illumination geometry) that would not transfer to a different laboratory [1].

12. What the Technology Cannot Prove Yet

• A thermal signal is not a unique chemical identification — thermography cannot confirm the Pb-Ba-Sb signature or any other elemental fingerprint.
• Classification within a known, closed set of firearms is not source attribution to an arbitrary unknown weapon.
• Laboratory conditions (fixed textile, fixed angle, fixed ammunition brand) are not crime-scene conditions.
• Spatial correspondence between model attention maps and expected residue zones is not proof the model has learned chemically meaningful features [1].
• Reported accuracy figures describe performance on this specific dataset and do not establish a universal or field-ready error rate.

Key Findings Box

Demonstrated: Flash-pulse IRT produces measurable thermal contrast associated with firearm-discharge deposition patterns on a single controlled textile; a multi-task deep-learning model trained on this data classified weapon category (94.3%), weapon model among five known options (85.1%), and ammunition type (92.6%), and estimated shooting distance with a mean absolute error of 7.92 cm, all under 5-fold cross-validation on 312 samples [1].

Suggested, not yet confirmed: That the thermal signal reflects genuine GSR chemistry rather than fabric damage, soot, or mechanical deformation alone; that model attention corresponds to physically meaningful residue zones (qualitative interpretability evidence only) [1].

Not yet established: Performance on dark, patterned, or synthetic textiles; performance at non-perpendicular angles; generalisation to firearms, ammunition, or cameras outside this dataset; casework validity; inter-laboratory reproducibility; any role as a confirmatory (rather than screening) method [1].

Limitations Box

Single textile substrate · single ammunition manufacturer and caliber · perpendicular shots only · small, imbalanced sample (n=312, 52% one firearm model) · no independent chemical confirmation of GSR on these samples · no inter-laboratory or cross-device testing · no environmental variability tested · no casework samples · qualitative-only interpretability analysis [1].

13. The Future: A Non-Destructive Layer of Forensic Ballistics?

The authors frame their own contribution modestly, as a foundation for future validation rather than a finished tool. Directions they explicitly identify include: expanding the textile and ammunition range (including non-toxic primers); testing variable shooting angles and environmental exposure; pairing thermal screening with SEM-EDX or Raman confirmation on the same sample; inter-laboratory reproducibility studies across different camera systems; and building uncertainty-aware models that flag low-confidence predictions rather than forcing a classification [1]. None of this is guaranteed to succeed — multi-task learning showed real trade-offs even within this single dataset, with distance-regression accuracy dropping when combined with classification tasks [1].

14. India: Could This Matter to Indian Forensic Laboratories?

SEM-EDS-based GSR research is active in India: a 2026 proof-of-concept study on GSR discrimination by SEM-EDX and chemometrics involved researchers at the National Forensic Sciences University (NFSU), Gandhinagar, the West Bengal National University of Juridical Sciences, and the Ballistics Division of the State Forensic Science Laboratory, Jaipur [12]. This confirms that Indian institutions are engaged in instrumental GSR characterisation broadly, and that SEM-EDS infrastructure and expertise exist within parts of the Indian forensic system (CFSL and State FSL networks, coordinated under the Directorate of Forensic Science Services).

No verified evidence was identified showing routine operational deployment of flash-pulse infrared thermography specifically in Indian forensic laboratories. The July 2026 study itself is a Czech laboratory investigation with no stated Indian institutional involvement, and no independent Indian replication of this specific method was located during research for this article. Any adoption in India would face practical barriers common to importing specialised NDE equipment — cooled mid-wave infrared cameras and calibrated flash systems are costly and require trained operators — on top of the same validation gaps the original authors flag for any laboratory.

15. From Seeing Residue to Measuring Evidence

The transformation this line of research is chasing is: visual evidence → measurable physical signal → quantified forensic information. The July 2026 study is scientifically interesting precisely because it treats that transformation as something to be tested, not assumed — its authors report both the successes (measurable classification accuracy, physically plausible attention patterns) and the gaps (no chemical confirmation, one textile, one ammunition line, no casework testing) in the same paper. Whether infrared thermography earns a routine place in forensic ballistic workflows now depends on exactly the kind of unglamorous work the authors call for next: more textiles, more firearms, more laboratories, and — eventually — real casework samples measured against a chemically confirmed ground truth.

Conclusion

Can forensic science turn an invisible physical response into a measurable evidentiary signal without destroying the original evidence? The July 2026 study offers a genuinely interesting, carefully hedged proof-of-concept answer: under one set of laboratory conditions, yes, to a measurable degree. It does not yet show that the signal is chemically specific to GSR, that it survives the diversity of real textiles and casework, or that it can stand anywhere but beside — never in place of — the confirmatory chemistry that forensic ballistics still relies on.

Frequently Asked Questions

Does infrared thermography identify the chemical composition of gunshot residue?

No. It detects thermal contrast believed to be associated with residue deposition, not elemental composition. Chemical confirmation still requires SEM-EDS or comparable analytical methods [1][3].

Is flash-pulse thermography destructive to the evidence?

No — it is a non-contact, non-destructive optical measurement that does not consume or alter the sample, which is one of its main proposed advantages over chemical colour tests [1].

Can this method determine the exact firearm used in a crime?

The study demonstrated classification among five specific, previously characterised firearm models with known reference data — not identification of an arbitrary unknown firearm from casework [1].

Was gunshot residue chemically confirmed in this study?

No. Ground truth relied on controlled firing conditions (known firearm, ammunition, and distance), not independent chemical confirmation via SEM-EDX or ICP-MS [1].

Does this technique replace SEM-EDS?

No. The study's own authors describe it as a complementary screening tool that could precede, but not replace, SEM-EDS confirmation [1][3].

Has this method been tested on dark or patterned fabrics?

Not in this study — only a single white bio-cotton substrate was used, despite dark-fabric visualisation being a commonly cited motivation for infrared-based GSR methods [1][4].

Is this technique used operationally in any forensic laboratory, including in India?

No verified evidence indicates routine operational deployment anywhere, including India. This is laboratory, proof-of-concept research [1].

What role does machine learning play in this research?

The July 2026 study used multi-task deep learning to classify thermal images, but the underlying physical measurement — flash-pulse infrared thermography — is a physics-based, non-AI technique that predates and is independent of the machine-learning analysis layered on top of it [1][9].

References

1. Sokol, M., Hejda, J., Volf, P., Valenta, M., Vávra, R., Kutílek, P. Preliminary evaluation of gunshot residue pattern analysis using flash-pulse infrared thermography and multi-task deep learning. Forensic Science International. 2026;384:112938. https://doi.org/10.1016/j.forsciint.2026.112938
2. Zeichner, A., Glattstein, B. Recent Developments in the Methods of Estimating Shooting Distance. The Scientific World Journal. 2002;2:573–585. https://doi.org/10.1100/tsw.2002.140
3. Goudsmits, E., Blakey, L.S., Chana, K., Sharples, G.P., Birkett, J.W. The analysis of organic and inorganic gunshot residue from a single sample. Forensic Science International. 2019;299:168–173. https://doi.org/10.1016/j.forsciint.2019.03.049
4. Yüksel, B., Ho, M., Ovide, O., Vander Pyl, C., Trejos, T. Infrared imaging as a complementary aid in estimating muzzle-to-target shooting distance: an application on dark, patterned and bloody samples. Turkiye Klinikleri Journal of Forensic Sciences and Legal Medicine. 2019;16:73–80.
5. Yüksel, B., Şen, N., Şeker, M.E., Öğünç, G.İ., Bulut, S. Integrating Chromophoric and Instrumental Methods (SEM-EDS and FTIR) for Accurate Shooting Distance Estimation: An Inter-laboratory Study in Forensic Chemistry. Current Analytical Chemistry. 2025;21:1333–1344.
6. Comparative SEM/EDS analysis of gunshot residue from non-toxic and traditional ammunitions employed by Dubai police. Frontiers in Chemistry. 2025. https://doi.org/10.3389/fchem.2025.1670901
7. Adav, S.S., Tan, Y.W.C., Low, C.T., et al. Exploring gunshot residue detection in fingerprints by functionalized particle-coupled matrix-assisted laser desorption/ionization mass spectrometry. Analyst. 2024;149:5704–5713. https://doi.org/10.1039/d4an01260g
8. Izraeli, E.S., Tsach, T., Levin, N. Optimizing FEG-SEM combined with an SDD EDX system for automated GSR analysis. X-Ray Spectrometry. 2013;43:29–37. https://doi.org/10.1002/xrs.2495
9. Moskovchenko, A., Švantner, M., Honner, M. Detection of gunshot residue by flash-pulse and long-pulse infrared thermography. Infrared Physics & Technology. 2024;140:105366. https://doi.org/10.1016/j.infrared.2024.105366
10. Švantner, M., Moskovchenko, A., Muzika, L., Skála, J., Honner, M. Quantitative Analysis of Flash-Pulse Thermographic Detection of Gunshot Residue. Proceedings. 2025;129:62. https://doi.org/10.3390/proceedings2025129062
11. Zain, Z.M., Jaluddin, S.N., Halim, M.I.A., Subri, M.S.M. The effect of type of firearm and shooting distance on pattern distribution, particle dispersion and amount of gunshot residue. Egyptian Journal of Forensic Sciences. 2021;11:1. https://doi.org/10.1186/s41935-021-00225-7
12. Harsheya, A., Das, T., Kumar, A., et al. Exploratory analysis of gunshot residue by SEM-EDX and chemometrics for firearm discrimination: a proof-of-concept approach. Australian Journal of Forensic Sciences. 2026. https://doi.org/10.1080/00450618.2026.2644190
13. Leiva Miranda, K.C., Ortega-Ojeda, F.E., García-Ruíz, C., Sáez Martínez, P. Shooting distance estimation based on gunshot residues analysed by XRD and multivariate analysis. Chemometrics and Intelligent Laboratory Systems. 2019;193:103831. https://doi.org/10.1016/j.chemolab.2019.103831
14. Shrivastava, P., Jain, V.K., Nagpal, S. Gunshot residue detection technologies — a review. Egyptian Journal of Forensic Sciences. 2021;11. https://doi.org/10.1186/s41935-021-00223-9
15. Sessa, F., Chisari, M., Esposito, M., et al. Advancing diagnostic tools in forensic science: The role of artificial intelligence in gunshot wound investigation — a systematic review. Forensic Sciences. 2025;5:30. https://doi.org/10.3390/forensicsci5030030