SECURITY SCREENING / TECHNICAL REVIEW

"Suspected Explosives / Drugs" Alerts from X-Ray Security Screening Machines —
How Credible Are They, Really?

From single-, dual- and multi-energy systems to the limits of the R value and the effective atomic number

Public literature + measured screenshot analysis | Nov 2025

Abstract At airports, metro stations and parcel-sorting centers, X-ray security screening machines routinely flag "suspected explosives" or "suspected drugs." Many people read such a flag as "the machine has identified a dangerous substance." In reality, mainstream screening machines perform X-ray attenuation measurements, not chemical composition analysis. A single-energy machine only sees "how much radiation was blocked"; a dual-energy machine uses the high/low-energy attenuation ratio R to sort materials into broad classes of organic / inorganic / metallic; a multi-energy machine adds more spectral channels, but the underlying physics is still attenuation. The effective atomic number Zeff is a collective average quantity, not a molecular fingerprint; the Zeff values of explosives and drugs overlap heavily with those of everyday items such as polyethylene, water, sugar and urea. Compounded by thickness, superposition, packaging, noise and the extremely low prevalence of threats, these "suspected" alerts generate large numbers of false alarms in real operations, carry low confidence, and cannot serve as a definitive conclusion. Drawing on public literature and measured screenshots, this article maps that capability boundary and offers actionable acceptance-testing and operational recommendations.

Keywords X-ray security screening; dual-energy; multi-energy; R value; effective atomic number; false alarm; explosives; drugs

1. Three Concepts First

The "energy" of a screening machine refers to the X-ray energy spectrum. The X-ray tube emits a continuous spectrum, and the detector receives the counts after the beam has penetrated the object. The attenuation of X-rays by matter follows an exponential law:

I = I₀ · exp[−∫ μ(E, l) dl]

where I₀ is the incident intensity, I is the transmitted intensity, μ is the linear attenuation coefficient, and the integral is taken along the ray path. μ depends on the material's atomic number, density and photon energy. The catch: a single measurement yields only one value of I, yet it contains two unknowns at once — "what the material is" and "how thick it is."

1.1 Single-Energy: A "Thickness Projection Image"

A single-energy machine has only one spectrum and outputs a grayscale image. The gray level reflects the integral of μ along the path — i.e., the product of density and thickness. It is good at revealing shape, contour and density anomalies, such as knives, guns, batteries and metal blocks; but it cannot separate a "thin high-density object" from a "thick low-density object." A thin steel plate and a thick plastic block can easily produce similar gray levels. Single-energy gives you a "projection," not a "composition."

1.2 Dual-Energy: Computing the R Value from Two Spectra

A dual-energy machine has two effective spectra: a low-energy one (e.g., the effective energy range corresponding to 80–120 kV) and a high-energy one (e.g., 140–160 kV or higher). In the low-energy spectrum the photoelectric effect dominates, making it more sensitive to atomic number; in the high-energy spectrum Compton scattering dominates, making it more sensitive to thickness. Dividing the logarithmic attenuations of the two images yields the R value:

R = ln(I_L / I_0L) ÷ ln(I_H / I_0H)

Under ideal conditions R correlates monotonically with the material's effective atomic number, so objects can be color-coded into three classes: orange for organic, green for inorganic, blue for metallic. That is the origin of the "orange food," "green glass" and "blue knives" seen on screening monitors. But R is an attenuation ratio — not a spectrum, still less a molecular fingerprint. It only answers "which class does it resemble," not "is it this specific substance."

1.3 Multi-Energy: More Channels, Same Physics

A multi-energy system has three or more energy bins; common implementations include dual-layer detectors, photon-counting detectors, kV switching or accelerator-based multi-energy imaging. The more energy channels there are, the more basis-material coefficients can be solved for, improving thickness compensation and material separation. But multi-energy is still sampling the attenuation coefficients: it adds "more equations," not "more chemical information." For organic compounds made of C, H, N and O, the attenuation curves are inherently very similar; multi-energy can deliver neither molecular structure nor elemental ratios.

Table 1 Single-, dual- and multi-energy systems compared

2. The R Value: Separates Classes, Not Substances

The physical basis of R-value classification is the difference in attenuation between elements at different energies. Low-atomic-number organic materials are dominated by Compton scattering and give higher R; high-atomic-number metals are dominated by the photoelectric effect and give lower R; inorganic materials fall in between. This ordering is robust at the class level, which is why dual-energy machines became ubiquitous.

But the R value is not a constant determined solely by the material. It is influenced by at least five factors:

• Thickness and beam hardening: the thicker the object, the more the low-energy component is absorbed, the harder the spectrum becomes, and the R value drifts; readings for the same material at different thicknesses can differ by a substantial amount.

• Superposition and occlusion: when two objects overlap, R is a blended value; a water bottle or a metal clasp is enough to skew the reading behind it.

• Noise and boundaries: low dose, edges, motion and scatter amplify the variance; the R value of a single pixel or a small region is especially unreliable.

• Mixtures and formulations: explosives contain binders and plasticizers, and drugs are often cut, diluted or turned into salts; change the formulation and both R and Zeff get flattened together.

• Calibration and machine-to-machine differences: spectra, filtration, detectors and basis-material decomposition algorithms differ between vendors, so R values cannot be compared directly across machines.

The correct use of the R value is therefore "class-level screening": sort objects into organic, inorganic and metallic, then judge in combination with shape, location, density and context. It cannot by itself answer "is this TNT" or "is this methamphetamine." Substances of the same class can span several R intervals, and substances of different classes can land in the same interval; move the threshold and false alarms and missed detections change at the same time.

This is why operators do not call "explosive" the moment they see orange. Orange only says "organic" — and water, sugar, bread, milk powder, cosmetics, tablets and plastic toys are all organic. The real judgment still relies on shape, packaging, image texture, opening the bag and chemical analysis.

Figure 1 Measured R-value distribution: focus on the overlap regions, not single points (original screenshots in D:\写作)

3. Zeff: An Average, Not an Identity Card

The effective atomic number Zeff is a quantity that "converts" a mixture into one equivalent atomic number. A common definition is:

Zeff = ( Σ w_i Z_i^m / Σ w_i Z_i^m−1 )^1/m, m ≈ 3

where w_i is the mass fraction of the i-th element and Z_i its atomic number. Zeff describes the overall attenuation behavior, not the molecular structure. Two completely different formulations will have similar Zeff as long as their C, H, O and N proportions are similar. That is the nature of an "average": it levels out differences.

A measured Zeff is not read out directly; it is "inverted" from the R value or from dual-energy basis-material decomposition. The entire chain includes: spectral stability, detector crosstalk and afterglow, thickness and density, calibration samples, the fitting algorithm and temperature drift. If any link is off, Zeff shifts. Typical uncertainties are on the order of ±0.5–1.5; thin objects, mixtures and superposition scenarios are worse.

More critically, the Zeff error is not random white noise but systematic: for the same material at different thicknesses, positions and packaging, the readings drift in swaths; different materials, in turn, can land in the same swath. Using single-point Zeff values for thresholding therefore looks physically justified but actually carries very low confidence.

Figure 2 Measured Zeff repeatability and spread across thicknesses (original screenshots in D:\写作)

4. Why Drugs and Explosives Cannot Be Separated from PE, Water and Sugar

The bulk of explosives and drugs consists of organic compounds of C, H, N and O. Explosives typically have densities of 1.2–1.9 g/cm³ and Zeff mostly in 6.5–8.2; drugs have lower densities and Zeff values, commonly 5.8–6.6. The problem is that common non-prohibited items fall squarely in the same range: polyethylene Zeff ≈ 5.7, water ≈ 7.4, sugar ≈ 6.5, urea ≈ 6.7, ammonium nitrate ≈ 7.4.

Table 2 Zeff and density of typical materials (shown to illustrate overlap, not as machine acceptance values)

Sorted by Zeff, TNT sits between sugar and ammonium nitrate; sorted by density, TNT is again right next to sugar. Neither axis provides separation. Set an "explosive window" of 6.5–7.5 and water, sugar, urea, ammonium nitrate and a great deal of food all fall inside; narrow it to 6.8–7.2 and TNT, RDX and PETN fall inside, but so do water, urea and some pharmaceuticals. No matter how the threshold is moved, there is no interval that "contains only dangerous items and no everyday ones."

Drugs are even more complicated. They are often diluted, adulterated, pressed into tablets or put into capsules, so Zeff is pulled in all directions by cutting agents and packaging; different locations in the same batch can even give different readings. Add liquids, powders, stacking and metal packaging, and the measured distribution widens further. Large within-class variance and small between-class gaps are the natural enemy of any classifier.

5. Low Prevalence Turns "High Accuracy" into Low Confidence

Even when a classifier performs well in the lab, in the real world its confidence is dragged down hard by one factor: the prevalence of dangerous items is extremely low. Look at the positive predictive value (PPV) through Bayes' formula:

PPV = Se·p ÷ [Se·p + (1−Sp)(1−p)]

Assume prevalence p = 1/10,000 and sensitivity Se = 90%. If specificity Sp = 95%, PPV is only 0.18%; at Sp = 99%, PPV is about 0.89%; at Sp = 99.9%, PPV is still only 8.26%. In other words, even with 99.9% specificity, every 10,000 normal items still produce about 10 false alarms, while the true positives number only 0.9. Most "suspected" flags on the screen are destined not to be dangerous items.

Table 3 PPV at low prevalence (p = 1/10,000, Se = 90%)

This is not a mathematical parlor game — it is one of the main sources of operational false alarms. On-site false alarms are further compounded by packaging, superposition, food, liquids, daily goods, electronics and metal parts, and then by operator alarm fatigue, throughput pressure and domain shift from new machines and new packaging. The end result: many alarms, very few items that genuinely need to be opened for re-inspection. Treat every alarm as "suspected explosive" and the lane grinds to a halt; let operators learn to ignore alarms, and the one truly dangerous case may slip through.

6. Conclusion: Downgrade "Suspected" to a "Re-inspection Trigger"

Putting the physics and the statistics together, the conclusion is simple: dual- and multi-energy X-ray imaging can separate objects into organic, inorganic and metallic classes, and can flag density and Zeff anomalies — but it cannot reliably identify explosives or drugs. Under current technology, a "suspected explosive / drug" alert is essentially a screening signal, not a definitive conclusion.

A more sensible operational practice therefore has three layers. Layer one: X-ray imaging looks at shape, location, density and Zeff anomalies and outputs "re-inspection recommended." Layer two: confirm with CT, XRD, ETD or by opening the bag. Layer three: feed the confirmation results back into the system to close the loop. When procuring and accepting machines, do not look only at AUC or laboratory accuracy — use PE, water, sugar, urea, ammonium nitrate, food, pharmaceuticals and cosmetics as a "confusion set," covering thickness, stacking and orientation; report the detection rate at a fixed false-alarm rate, report false alarms per thousand items, and report PPV at realistic prevalence.

The wording must be downgraded too: output "anomaly / re-inspection recommended," not "explosive / drug"; keep the original image and the re-inspection record so people can see why the alarm fired. Multi-modality, thickness compensation and confidence calibration are the right R&D directions, but "suspected" must not be repackaged as "confirmed." Under the current physics, "suspected" has screening value but no definitive identification value.

None of this denies the value of screening machines. They remain irreplaceable for shape, density, prohibited-item contours, liquids and metal detection. What needs correcting is the narrative that misreads "material class + anomaly score" as "composition identification." Drawing the capability boundary clearly is precisely what lets the equipment be used in the right place.

Note on the measured screenshots: the screenshots used in this article come from the author's own measurement records, including R / Zeff readings for different materials, thicknesses and placements. Because the screenshots do not annotate every machine parameter, the main text discusses only distributions and overlaps and never cites a single reading as a conclusion. The original screenshots are stored in D:\写作 and can be viewed in the companion PPT.

References

[1] NIST. X-Ray Mass Attenuation Coefficients. https://physics.nist.gov/PhysRefData/Xcom/html/xcom1.html

[2] Fraunhofer Publica. X-ray-absorption measurements of explosives and the effective atomic number Zeff. https://publica.fraunhofer.de/bitstreams/c01fcefc-9b01-4fa8-a770-595b67502bf1/download

[3] IAEA INIS. High accuracy X-ray dual-energy experiments and non-rotational tomography algorithm for explosives detection technique in luggage control. https://inis.iaea.org/records/h2cxz-evj69

[4] SPIE, 2025. Enhancing Security Measures: Leveraging Fuzzy Logic Integration for Prohibited Item Detection in Dual-Energy X-Ray Security Inspection. https://web.cs.hacettepe.edu.tr/~onderefe/PDF/2025spie.pdf

[5] ScienceDirect. Comparative investigation of photon interaction parameters in psychoactive substances: the role of molecular composition. https://www.sciencedirect.com/science/article/pii/S2211379726001907

Note: the Zeff and density values in this article are typical values from public literature / commonly used calculation conventions, used to illustrate material overlap relationships; they do not serve as acceptance values for any specific machine or algorithm. The measured screenshots are used only to show distributions and repeatability; refer to the original records for the specific machines and parameters.