BloodNet: An attention-based deep network for accurate, efficient, and costless bloodstain time since deposition inference.

This article focuses on BloodNet, a novel deep learning-based method for inferring the time since deposition (TSD) of bloodstains using macroscopic bloodstain photographs. Developed with a large-scale benchmark database of approximately 50,000 bloodstain images from rabbit samples, BloodNet employs attention mechanisms within a convolutional neural network to capture fine-grained visual features that correlate with bloodstain aging, achieving high accuracy and generalizability across validation and external test sets. The study also presents a paired microscopic analysis using Raman spectroscopy combined with machine learning, which, while effective, demonstrated lower accuracy than BloodNet, highlighting the potential of macroscopic deep learning approaches for practical forensic bioinformatics applications. Limitations include the use of rabbit blood and laboratory conditions, suggesting the need for further validation with human samples and real crime scene data.