Rapid forensic differentiation of human and animal bones using handheld near-infrared spectroscopy and deep learning.

The forensic differentiation of human and animal bones is critical in various investigations, mainly when dealing with fragmented skeletal remains. This study explores the practical application of handheld near-infrared spectroscopy combined with artificial neural networks for rapid, non-destructive identification of bone species. 225 femoral bone samples, including human and animal specimens, were analyzed using near-infrared spectrometry. Preprocessed spectral data were classified using binary and multi-class artificial neural networks models, achieving high median accuracy rates of 96.3 % and 77.8 %, respectively. Principal component analysis revealed clustering patterns across species and revealed the influence of post-mortem intervals on human bone spectra. The study demonstrates the practicality of handheld near-infrared devices for on-site forensic applications, offering a portable and efficient alternative to traditional destructive methods and the potential for further refinement of this technology to achieve significant advancements in forensic science, particularly in species determination and post-mortem analyses.