利用卷积神经网络对毛发物证显微图像进行自动分类,为进一步提高显微检验技术的自动化程度和毛发物证检验提供技术参考。采用徕卡DVM6数码显微镜在1400倍放大条件下采集6类毛发共60000张样本图像,构建毛发分类数据集。基于卷积神经网络搭建Hair-Net模型,通过该模型对毛发分类数据集进行样本训练和测试验证。实验研究表明,经过参数调试和优化手段的改进后,新的Hair-Net分类精度最高可达97.82%,成功实现了毛发物证显微图像的自动分类,增强了稳健性。

This paper uses a convolutional neural network to automatically classify microscopic images of hair evidence with the aim of enhancing the automation of microscopic technology and providing technical reference for test efficiency. Six kinds of microscopic hair images are collected via Leica DVM6 microscope and are magnified 1400 times to form the sample image dataset which contains 60000 images. The network model Hair-Net based on the convolutional neural network is used to conduct sample training and testing using different parameters. Experimental results show that the classification accuracy of improved Hair-Net can reach 97.82% after parameter testing and optimization, demonstrating that this method can realize automatic classification of microscopic hair images and enhance the robustness.