Indian Journal of Science and Technology

Abstract

Objectives: To maximize the identification and improve the accuracy of classifying the bloodstain in a hyperspectral image (HSI) at the crime scene, a 3-D discrete wavelet transform (3-D DWT) Dense CNN deep learning model is proposed in this work. Methods: This work proposes the use of a 3-D DWT to pre-process HSI data to effectively extract both spatial and spectral information while maintaining robust feature representation capabilities. Then, 3-D CNN that integrates dense connections attaches great importance to the reuse of features for classification. The experiment was carried out with the initial training/testing ratio set to 10/90 of the data samples, and we compared the results with four different state-of-the-art CNN architectures. Findings: The experimental results show that the 3-D DWT Dense CNN deep learning model achieves 97% classification accuracy, smoother classification maps, and more discriminable features for hyperspectral image classification. Novelty and Application: This work provides a deep learning 3D dense CNN model with the 3-D DWT and achieves improved identification of bloodstains at a messy crime scene. The proposed model requires a smaller number of trainable parameters, less computational power, so it can be used in the field of forensic science, where substance classification at the scene is important.

Keywords: Hyperspectral Imaging, Blood Strain Classification, Discrete Wavelet Transform (DWT), 3­D CNN, Dense Connection, Forensic Science