Infrared handprint image is an image that applies infrared imaging technology to criminal investigation and other special scenes. It can be used to detect traces that cannot be directly observed under visible light.Efficient identification and analysis of handprint are conducive to obtaining more information for solving cases. However, due to thermal diffusion, the depth fuzzy feature of infrared handprint is not conducive to detection and classification, and the convolution neural network technology is widely used in the field of natural image classification because of its excellent feature extraction ability.Therefore, aiming at the problem of fuzzy infrared handprint classification, we design a novel convolution neural network, which includes a convolutional layer, small MBConv block and fully connected layer.We choose EfficientNet which is suitable for infrared handprint classification from classical convolution neural network as our basic network. And propose a small MBConv block to improve the network model, so that the network has fewer training parameters, effectively reduces the problem of over fitting, and improves the classification performance compared with the original model.We use our model for the automatic classification of infrared handprint images. The results show that our model achieves the average accuracyto 95.78% for multi-class classification, which is 2.19% higher than the original model.

红外手印图像是将红外成像技术应用于刑事侦查和其他特殊场景的图像。它可以用来检测在可见光下无法直接观察到的痕迹。有效的手印识别和分析有助于获得更多的信息来解决案件。然而，由于热扩散，红外手印的深度模糊特征不利于检测和分类，而卷积神经网络技术由于其出色的特征提取能力而在自然图像分类领域得到了广泛的应用。针对模糊红外手印分类问题，设计了一种新颖的卷积神经网络，包括卷积层，小MBConv块和全连接层。我们从经典卷积神经网络中选择适用于红外手印分类的EfficientNet作为我们的基本网络。并提出了一个较小的MBConv块来改进网络模型，使网络具有较少的训练参数，与原始模型相比有效减少了过度拟合的问题，并提高了分类性能。手印图像。结果表明，我们的模型对多类分类的平均准确率达到95.78％，比原始模型提高了2.19％。与原始模型相比，有效地减少了过度拟合的问题，并提高了分类性能。我们使用该模型对红外手印图像进行自动分类。结果表明，我们的模型对多类分类的平均准确率达到95.78％，比原始模型提高了2.19％。与原始模型相比，有效地减少了过度拟合的问题，并提高了分类性能。我们使用该模型对红外手印图像进行自动分类。结果表明，我们的模型对多类分类的平均准确率达到95.78％，比原始模型提高了2.19％。