深度可分离卷积神经网络miniXception对矿工情绪特征的识别  

Miner’s emotion recognition based on deepwise separableconvolution neural network miniXception

作　　者：王征[1] 张科 张赫林 潘红光[1] WANG Zheng;ZHANG Ke;ZHANG Helin;PAN Hongguang(College of Electrical and Control Engineering,Xi’an University of Science and Technology,Xi’an 710054,China)

机构地区：[1]西安科技大学电气与控制工程学院,陕西西安710054

出　　处：《西安科技大学学报》2022年第3期562-571,共10页Journal of Xi’an University of Science and Technology

基　　金：国家自然科学基金项目(51804249)。

摘　　要：为准确了解煤矿井下矿工情绪状况,以陕西省某煤矿为研究区,选取并建立矿工表情图像数据集。基于深度可分离卷积神经网络miniXception搭建矿工表情识别模型,对其残差连接进行改进,加入多次标准卷积与轻量化上下采样模块,并提出Exp-FReLU作为网络主分支的激活函数。通过MMA面部表情公共数据集及文中自制数据集对网络进行训练,输出每类表情的识别率并将识别率最高的分类结果视作预测结果。实验分析了训练时间、精确度、召回率、F1分数以及分类准确度混淆矩阵,发现改进miniXception网络对生气、厌恶、恐惧、高兴、沮丧、惊讶以及中性7种表情的识别率分别为86%,76%,67%,97%,63%,88%以及72%;经过100次迭代,模型总体准确率达到0.833,损失值最低降至0.086。研究表明,改进miniXception网络在矿工面部表情的识别问题上具有可行性,能够满足实际应用需要。In order to learn more about the miners’expression in coal mines,the image set about miner’s emotion is established from a coal mine in Shaanxi.Based on miniXception,a deepwise separable convolution neural network,a miner expression recognition model is constructed.Its residual connection is improved,in which several standard convolutions and lightweight upsampling and downsampling modules are added,and Exp-FReLU is applied as an activation function in the backbone.With the MMA,a facial expression common dataset,and our self-made dataset,the network is trained to capture the recognition rate of different expressions,and the result with the highest recognition rate is regarded as the prediction result.The training time,precision,recall,F1 score and classification accuracy confusion matrix are analyzed experimentally.It is found that the recognition rates of the seven emotion improved by miniXception are 86%,76%,67%,97%,63%,88%and 72%for anger,disgust,scare,happiness,sadness,surprise and neutral expression,respectively.After 100 iterations,the overall accuracy of the model is 0.833,and the loss is as low as 0.086.The research results indicate that the improved miniXception is feasible in classifying the miner facial expression with the practical application satisfied.

关 键 词：深度学习 矿工面部表情识别 表情特征提取 深度可分离卷积 miniXception 

分 类 号：TD76[矿业工程—矿井通风与安全]