Age estimation of bloodstains based on convolutional neural network algorithm and hyperspectral imaging technology

Abstract

The most common trace of evidence at a crime scene is blood. In judicial scientific proof, it is regarded as one of the most important material types of evidence. In this study, the spectral data of the bloodstain samples with different time were collected via a hyperspectral imaging system with a spectral range of 400–1000 nm and spectral sampling interval of 5 nm. The training model of blood aging was built using the Convolutional Neural Network (CNN) algorithm and the above hyperspectral spectral data. It was also compared with the traditional partial least squares (PLS) and extreme learning machine (ELM) models. The experimental results showed that the performance of CNN model was the best, with a determination coefficient (R²) of 0.987, which is higher compared to that of the PLS model (R² = 0.883) and ELM model (R² = 0.936). Besides, the root mean square error of prediction (RMSEP) of the CNN model was 6.949 hours, smaller than the PLS model (RMSEP = 18.752 hours) and the ELM model (RMSEP = 13.717 hours). The mean absolute percentage error (MAPE) of the prediction set was 0.49% when using the CNN algorithm, which was also the minimum of the three algorithms, and it represented that the prediction results were the best. The experimental data showed that the method proposed in this study could accurately estimate the age of bloodstains, providing a new technology reference for bloodstain detection.