Deep learning method based on image recognition for intra-puparial age and postmortem interval estimation in the...

Accurate estimation of intra-puparial age in necrophagous flies is essential for determining the postmortem interval (PMI) in forensic entomology. Traditional methods based on morphological observation of intra-pupal structures are widely used but rely on complex diagnostic criteria and are subject to observer bias, posing a technical bottleneck in PMI estimation using insect evidence. Deep learning, particularly image-based methods, offers a promising solution for objective and automated identification in forensic entomology. Sarcophaga peregrina (Robineau-Desvoidy, 1830) (Diptera: Sarcophagidae) is a common necrophagous fly species. In this study, we propose an image-based deep learning framework for automatic classification of intra-pupal developmental age in S. peregrina to enhance the accuracy of PMI estimation. Pupae were reared at 25 °C, and samples from different developmental stages (Day 1 to Day 11) were collected. After removing the puparium, high-resolution images of intra-pupal morphology were captured to construct a dataset. A ResNet50 network was first employed to extract regions of interest, followed by a Vision Transformer (ViT) model for end-to-end classification of developmental stages. The proposed method achieved a classification precision of 94.00 %, recall of 93.41 %, and F1-score of 93.43 %. These findings demonstrate that deep learning can serve as an effective and objective alternative to manual morphological assessment, reducing reliance on expert experience in intra-puparial age estimation. The proposed approach establishes a viable AI-assisted pathway for standardized, rapid, and accurate PMI inference based on insect evidence, offering practical value for forensic investigations.