BloodNet

BloodNet infers the time since deposition (TSD) of bloodstains from macroscopic photographs using attention-based deep learning to support forensic bloodstain aging analysis.

Key Features:

• Attention Mechanisms: Employs attention-based mechanisms within a deep neural network to focus on localized fine-grained features in high-resolution bloodstain images.
• Macroscopic Analysis: Operates on macroscopic photographs of bloodstains rather than relying on microscopic or spectroscopic measurements.
• Large-Scale Benchmark Database: Trained and evaluated on a benchmark dataset of approximately 50,000 bloodstain photographs with varying TSDs.
• Visual Analysis Tools: Uses visualization methods such as Smooth Grad-CAM to demonstrate learned local patterns associated with specific TSDs.
• Comparative Performance: Reported to outperform a paired microscopic approach based on Raman spectroscopy and machine learning with Bayesian optimization in accuracy for TSD inference.

Scientific Applications:

• Forensic time-since-deposition estimation: Determines TSD of bloodstains to provide temporal information relevant to crime scene investigations.
• Non-destructive photographic workflows: Enables age estimation from standard photographic images, supporting scene-level or field analyses without spectroscopic sampling.

Methodology:

Training an attention-based deep neural network on a large-scale dataset of bloodstain images and interpreting learned attention patterns with Smooth Grad-CAM.