Analysis of passive bloodstain morphology across surface textures and drop heights using deep learning

Bloodstain pattern analysis (BPA) is a critical forensic science tool for reconstructing crime scene events. In this study, the effect of substrate type and drop height on the morphology of passive bloodstains was examined under controlled laboratory conditions. Blood samples were dropped vertically at 90° angle from three different heights, and the drops were permitted to strike five different surfaces, including curved cups, crushed chart paper, jute cloth, jelly stone, and concrete. These substrates were chosen to represent a realistic range of porous, semi-porous, non-porous, textured, and curved materials that are commonly encountered in crime scenes. The features of the substrate affect stain morphology, including shape irregularity and satellite formation, but not the measured angle of impact. These findings validate the consistency of impact angle determination using BPA, wherein the nature of the substrate primarily affects stain morphology but not necessarily the accuracy of angles. The large image data sets were tested using deep learning approaches, which effectively differentiate bloodstain patterns generated from varying fall heights. MobileNet model, leveraging pretrained ImageNet features, achieved superior accuracy and generalisation, underscoring the value of transfer learning for small forensic datasets. Future extensions of this work will include multiple impact angles, motion-related effects and temperature-controlled conditions to represent the actual crime scene scenarios. Deep learning-based analysis of these data may improve the understanding of bloodstain morphology and strengthen the forensic applicability.