Forensic applications of hyperspectral imaging technique: a narrative review

3.1 Bloodstain analysis

Of all biological fluids, blood is one of the commonest to be detected at a crime scene. The interpretation of the bloodstain features, at macroscopic (shape, size, distribution, colour) and microscopic-biochemical level, represents one of the cornerstones of the forensic investigation process.

The tests commonly used are compliant with the chemical methods such as Kastle-Meyer (KM), leuco-malachite green (LMG), benzidine and Luminol. For all the above-mentioned methods, although characterised by several false positive results (Luminol is the most reliable), confirmation tests such as spectroscopic and/or chromatographic analysis are required. One limitation of these methods is the partial destruction of samples. The use of a method, such as HSI, where no contact is required between samples and instruments avoids both the dispersion of the sample and its contamination. Oxyhemoglobin and methemoglobin perform characteristic peaks in the visible light spectrum, marked as β (≈542 nm) and α (≈576 nm), which detect blood and differentiate it from other substances. ⁵

Zulfiqar et al. ⁶ proposed an HSI-based experimental protocol able to differentiate bloodstains from other substances similar in colour and consistency (ketchup, rust-coloured acrylic paint, red and brown acrylic paint, red and rust nail polish, fake blood and red ink) on different substrates (white cotton fabric, white tiles and PVC sheet); besides, the samples were examined for three days to investigate their behaviour over time. The authors concluded that the proposed method achieved the best results.

As pointed out by Romaszewski et al., ⁷ HSI applicability may be impaired by some factors, such as the effect resulting from the bloodstain spectral mixing with the underlying surface, the time-related bloodstains spectra changes, a possible mismatch between the reference spectrum and the target spectrum, and differences in acquisition equipment, scene and lighting complexity. To address the above-mentioned potential limitations and generate algorithms for the enhancement of bloodstain recognition, the same authors created a representative patterns-based dataset configured according to a series of predetermined scenarios. Again, images were acquired at different time frames for no more than three weeks in order to detect time-related changes.

Other authors, ¹ have applied some machine learning models-based functions to categorise the images processed by the aforementioned dataset, exploiting the potential of both the approaches synergistically.

Crowther et al., in their recently published survey, ⁸ highlighted how hugely helpful HSI may be in the detection and positive identification of bloodstained footwear marks, using both undiluted and diluted with distilled water blood material at various dilutions (1: 5, 1:10, 1:20, 1:50) applied to a range of different substrates and colours. The results showed HSI as the gold-standard in detecting traces left on white tiles, laminate, carpet and blue tiles compared with the digital photography technique, which succeeded only on black tiles. In particular, the HSI method showed a percentage greater than 75% of footwear evidence on white tiles at all dilutions. Similarly, Cadd et al. investigated the possibility of taking fingerprints, using blood material at different dilutions up to 6 months apart. ⁹ With regard to time-related bloodstains changes, some authors have documented the hyperspectral imaging capacity to detect traces left up to 200 days before, whose absolute value margin of error increases with the increasing time-frame, achieving, as in this case, important results. ¹⁰

3.2 Gunshot residues analysis

Another crucial matter for forensic examiners carrying out a crime scene investigation, where a firearm has been used, is the detection of gunshot residues (GSR), that may often be invisible to the naked eye.

The particles formed by firing are mostly composed of heavy metals depending on the different types of ammunition. The inorganic particles (IGSR) examined can remain in a normal environment indefinitely and although real particles under normal conditions do not degrade, those deposited for example on an individual’s hands or clothing may get lost over time, because of an individual’s movements. In this case too, some validated and existing methods, such as scanning electron microscopy with energy dispersive X-ray spectroscopy (SEM-EDX), are used to detect the presence of deposited particles on different surfaces.

With regard to cartridge discharge residues, using similar bloodstain methodology to that used by other authors, Glomb et al. studied the possibility of identifying inorganic particles deposited on different textiles combining hyperspectral imaging with machine learning analysis algorithms. They concluded that an accurate acquisition is certainly possible.

Other authors used hyperspectral microscopy (ATR FT-IR) to detect organic gunshot residues (OGSR) deposited on the shooter's skin. ¹² It is widely acknowledged that cartridge discharge residues also consist of organic compounds (containing nitrocellulose, nitroglycerin and nitroguanidine) mostly resulting from the propellant and the primer particle that, if detected, can be used to prove the explosion originated from a shot. One of the main advantages of using these molecules, compared with metal residues, is that the latter are widely distributed in the living and working environment, potentially causing false positive results. Existing methods used for the gunshot residues investigation are gas chromatography (GC) and high-performance liquid chromatography (HPLC) together with mass spectrometry (MS). ¹³ The results from the above suggest that HSI provides a complementary method in conjunction with conventional detection techniques.

3.3 Fingerprint analysis

The analysis and detection of fingerprints where a crime has occurred or is suspected are two of the most relevant and sensitive steps in an investigation. The Study of fingerprints in resolving legal cases is even more decisive than DNA analysis. It has been reported that fingerprint analysis helps solve about 10 times more cases than DNA analysis. ¹⁴ Fingerprints are a biometric characteristic unique to each individual, even twins, that remain unchanged even with ageing. A recent review of the scientific literature indicates that the most common method is mass spectrometry used alone to identify the constituents of fingerprints, or together with chromatographic methods, to provide accurate results in terms of fatty acids present in fingerprints. ¹⁴ It is in fact known that fingerprints consist of a mixture of substances such as amino acids (serine is the most commonly detected), proteins and fatty acids; it is precisely the presence of these substances that allows an identification even of latent fingerprints macroscopically invisible to the naked eye.

With regard to the use of HSI, the capacity to detect latent fingerprints, that otherwise, could remain undetected, appears particularly promising. Akiba et al. ¹⁵ state it is also possible to detect and analyse latent and still untreated fingerprints using traditional chemical methods (used in order to increase the detection method responsiveness by enhancing the contrast of fingerprints with respect to the surface on which they were imprinted). In their study, the authors used a portable hyperspectral imager combined with a continuous wave green laser.

3.4 Documents analysis

Document analysis in the forensic field usually aims to identify fraud and adulteration. Further, like fingerprints, handwriting analysis is useful for identification purposes. Existing conventional methods applied in this area require a contact with evidence and, in some cases, even their destruction. Therefore, the introduction of HSI would offer an undoubted advantage, especially in those cases where documents preservation is essential for judicial purposes.

One of the main practical HSI limitations in this context is the low availability of a reference dataset. Starting from this assumption in the work published by Islam et al. ¹⁶ the authors introduced a dataset of hyperspectral images about handwritten documents drawn up by 54 individuals, whose age and gender were known. Each person was required to write 28 sentences, containing upper and lower-case letters using 12 different types of pens (all blue in colour). At the end of the collection step, The documents were then analysed using a hyperspectral cameral obtaining a total of 270 images.

With regard to its practical application, Silva et al. ¹⁷ published a paper to propose HSI-NIR (near infrared) use combined with multivariate analysis in order to identify three types of counterfeiting (deletion and addition of both text and lines intersecting the original text). Although some evidence remains undetected, the authors concluded positively supporting the enormous potential in forged documents detection.

Melit Devassy ¹⁸ instead proposed a hyperspectral images dataset analysis concerning 40 different types of paper using an algorithm (t-Distributed Stochastic Neighbour Embedding algorithm) showing an excellent discriminative performance compared to the Principal Component Analysis (PCA) method considered by the same authors as standard.

3.5 Further forensic applications

Kalacska et al. ¹⁹ investigated HSI performance in detecting burial sites and mass graves. This type of forensic investigation assumes great importance where it has been calculated hundreds of thousands of people have died as a result of wars or feud among criminal gangs and been clandestinely buried in mass graves. The potential of using new integrated technologies such as HSI opens the possibility of using them without actually inspecting the site but by using aerial image recognition with the aim of minimising risks to the forensic examiners’ safety.

In this experimental study, the authors reproduced, under known humidity and temperature conditions, a 5 m² sized burial site where animal carcases were buried and observed over a period of 16 months. By the comparison with another site, identical in structure but filled with nothing but soil, it was highlighted that in the site where corpses were buried, typical substances, including but not limited to dimethyl disulfide, toluene, benzene, cadaverine and putrescine, ²⁰ evolved from the decaying human body processes. The presence of specific substances is exactly the principle on which some methods such as HSI are used. The authors showed how this method performed in detecting burial pits via aerial hyperspectral images acquisition.

It should be emphasised that the substances evolving from a buried decaying corpse (especially nitrogen), may affect chlorophyll concentrations in the plants growing over the pit as a result of being fed by the aforementioned substances. ²¹ With this assumption in mind, Silvan-Cardenas et al. ²¹ used 10 farm pig carcases to reproduce a burial site, burying the animals at depths ranging from 1.1 to 1.5 metres. The aerial hyperspectral images acquired over variable time ranges showed some hyperspectral indices actually useable in order to detect nitrogen-enrichment present in plants.

Another interesting application is proposed by Payne et al. ²² in which HSI was used, in comparison with spectrophotometry and simple photography, to estimate the age of bruises. In the reported experimental work, the authors had injected blood together with a mixture of blood and bile at different dilutions into the subcutaneous tissue of a pig, highlighting the superiority of hyperspectral imaging performance compared with spectrophotometry.

In their latest published study, ²³ Melit Devassy et al. investigated hyperspectral imaging applicability for the analysis of 12 different types of commonly used alcoholic and non-alcoholic beverages applied onto 4 different types of paper towels, in order to assess samples' behaviour over time.

A further particularly significant application is the possibility of identifying drugs and substance abuse. As reported by some authors, ²⁴ HSI-NIR combined with machine learning technology have shown considerable potential in identifying Cannabis sativa Linnaeus leaves samples.