基于深度学习与SVM的吸毒成瘾者识别

Identification of Drug Addicts by Deep Learning and SVM

提出一种使用PCA和线性判别器的神经网络模型,利用深度学习方法通过面部图像及抽象特征中的局部信息识别吸毒成瘾者的成瘾程度和社区矫正时间。首先对主干网络Res Net50进行预训练;再使用PCA降低特征数、Fisher判别器进行预判,从而使模型的训练时间减少、特征提取更加准确和快捷;最后网络末端通过全连接层与SVM函数的组合进行分类。随机梯度下降的优化目标采用了交叉熵损失。实验表明,此方法对于吸毒成瘾程度的识别准确度可达81.74%,对于社区矫正时间的识别准确率可达60.59%。

A neural network model using PCA and linear discriminator is proposed to identify addiction level and community correction time of drug addicts according to facial images as well as local information in abstract features through deep learning method.Firstly,the backbone network ResNet50 is pre-trained.Then,PCA is used to reduce the number of features and the Fisher discriminator is used for pre-discrimination,so that,the training time of the model is reduced and feature extraction is more accurate and faster.Finally,the net-work end is classified through the combination of the fully connected layer and the SVM function.Cross-entropy loss is adopted as the optimization goal of stochastic gradient descent.The experimental results show that the method has a recognition accuracy of 81.74%for the degree of drug addiction and 60.59%for the community correction time.