公共服务窗口一直是政府为民众服务的重要渠道，因而窗口服务态度的好坏受到社会各界的广泛关注。传统的窗口服务态度评定方法主要有人工记录、调查问卷等方式，这些方式不仅会耗费大量的时间与人力，而且采集到的信息往往不够全面、准确度也不高，极易受到人的主观意识影响，造成不公。近年来，深度学习方法在目标检测和识别方面取得了较好的效果，人脸表情识别作为图像识别领域的重要分支，表情识别研究也取得了很大进展。将表情识别应用到服务窗口态度评判中，可以更准确、客观地对服务态度进行定量评价和分析。但是现有的表情识别方法应用到服务窗口表情识别中时，检测准确率和时效性较低，同时缺少利用识别出来的单项表情进行服务态度的综合评价方法。本文基于深度学习最新研究成果，结合服务窗口表情识别任务特定场景，实现对服务窗口人脸表情的准确、快速识别。同时提出服务窗口服务态度综合评判模型，根据服务窗口表情识别结果对服务态度进行分级评价。本文研究方法对准确识别服务窗口的表情、改善窗口服务态度具有重要意义。本文的主要工作及成果如下：

       （1）提出了基于改进SSD的服务窗口人脸检测算法。本文分析目前主流的目标检测算法，选择基于端到端的SSD算法作为基础。由于待检测的服务窗口人脸可能会出现几何形变问题，因此在SSD算法的基础网络部分加入可变形卷积模块，使得卷积层拥有自适应感受野，提高对服务窗口人脸形状的敏感性，从而提高服务窗口人脸检测精度。同时对网络模型进行三元量化，将神经网络中的权重参数降低为三元值，减小模型规模，提高网络运算速度，解决神经网络庞大的计算量和存储开销问题，提高其在小型设备上的适用性。通过实验对比SSD、DCN-SSD和三元量化DCN-SSD算法在公共服务窗口人脸检测数据集上的检测效果，证明改进后的算法在人脸检测精度和速度上效果都更好。

       （2）提出了基于改进胶囊网络算法的服务窗口表情识别算法。首先针对胶囊网络中特征提取层只有一层卷积层，特征提取不充分的问题，使用VGG16网络代替胶囊网络的原特征提取层，变换后的胶囊网络特征提取阶段具有13层卷积结构，可以有效提取深层抽象特征，获取更多的空间特征信息，提高表情识别的准确率。然后将条件生成对抗网络和改进后的胶囊网络进行融合，实现对侧脸表情的识别。融合后的网络先利用条件生成对抗网络中的生成器进行浅层特征提取，然后把提取到的特征信息输入到VGG16网络中，成为了一个端到端的中层特征提取、分类网络。网络的浅层部分实现的是侧脸表情到正脸表情的映射，高层部分则是对正脸表情的判别。实验结果表明改进了的CGAN-CapsNet算法在表情识别的准确率上要远高于原胶囊网络，同时CGAN-CapsNet算法还实现了对侧脸表情图像的识别。

（3）提出了一种服务窗口服务态度综合评判模型。首先将基于三元量化的DCN-SSD人脸检测算法与基于CGAN-CapsNet的表情识别算法结合起来构建了服务窗口表情识别模型。然后根据服务窗口的表情识别结果，总结了服务窗口服务态度综合评判指标，通过主观赋权法为这些指标赋予不同的权重。最后基于TOPSIS法设计了服务窗口服务态度综合评判模型，并进行了实验验证。验证结果表明，该综合评判模型依据服务窗口表情识别结果可以客观、准确地对服务窗口的服务态度进行评判。

[1] 陈雅檀. 公共服务窗口 的服务伦理问题研究[D].湖南农业大学,2016.

[2] 李娥.人脸检测方法综述[J].信息技术与信息化,2018(04):24-26.

[3] Li S Z, Zhu L. Zhang Z, et al. Statistical Learning of Multi-view Face Detection[C].european conference on computer vision, 2002: 67-8I.

[4] Viola P A , Jones M J . Rapid Object Detection using a Boosted Cascade of Simple Features[C]// IEEE Computer Society Conference on Computer Vision & Pattern Recognition. IEEE, 2001.

[5] Mathias M, Benenson R, Pedersoli M, et al. Face Detection without Bells and Whistles[J]. european conference on computer vision., 2014: 720-735.

[6] Krizhevsky A , Sutskever I , Hinton G . ImageNet Classification with Deep Convolutional Neural Networks[J]. Advances in neural information processing systems, 2012, 25(2).

[7] Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]. // roceedings of the IEEE conference on computer vision and pattern recognition. 2014:580-587.

[8] Zhang Z . [IEEE Proceedings of the Seventh IEEE International Conference on Computer Vision - Kerkyra, Greece (1999.09.20-1999.09.27)] Proceedings of the Seventh IEEE International Conference on Computer Vision - Flexible camera calibration by viewing a plane from unknown orientations[J]. 1999:666-673 vol.1.

[9] Ren S, He K, Girshick R, et al. Faster r-cnn: Towards real-time object detection with region proposal networks[C]. // Advances inneural information processing systems. 2015:91-99.

[10] Liu W , Anguelov D , Erhan D , et al. SSD: Single Shot MultiBox Detector[J]. 2016.

[11] Rowley H A , Baluja S , Kanade T . Rotation Invariant Neural Network-Based Face Detection[C]// Computer Vision and Pattern Recognition, 1998. Proceedings. 1998 IEEE Computer Society Conference on. IEEE, 1998.

[12] Zhang K , Zhang Z , Li Z , et al. Joint Face Detection and Alignment Using Multitask Cascaded Convolutional Networks[J]. IEEE Signal Processing Letters, 2016, 23(10):1499-1503.

[13] Yang S, Luo P, Loy C C, et al. Faceness-Net: Face Detection through Deep Facial Part Responses[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2018, 40(8): 1845一1859.

[14] 周慧娟,张强,刘羽,王旭阳,柳颖.基于YOLO2的地铁进站客流人脸检测方法[J].计算机与现代化,2019(10):76-82.

[15] 刘雪燕,李明.基于深度学习混合模型的人脸检测算法[J].信息与电脑(理论版),2019,31(20):38-41.

[16] Wan S, Aggarwal J K. A scalable metric learning-based voting method for expression recognition[M].2013.

[17] Ali H, Hariharan M, Yaacob S, et al. Facial emotion recognition using empirical mode decomposition[J]. Expert Systems with Applications An International Journal, 2015, 42 (3):1261-1277.

[18] Aleksic P S, Katsaggelos A K. Automatic facial expression recognition using facial animation parameters and multistream HMMs[J]. IEEE Transactions on Information Forensics and Security, 2006, 1(1): 3-11.

[19] Fan Y, Lu X, Li D, et al. Video-based emotion recognition using CNN-RNN and C3D hybrid networks[C]//Proceedings of the 18th ACM International Conference on Multimodal Interaction. 2016: 445-450.

[20] Kim B K, Lee H, Roh J, et al. Hierarchical Committee of Deep CNNs with Exponentially-Weighted Decision Fusion for Static Facial Expression Recognition[C]// ACM on International Conference on Multimodal Interaction.ACM, 2015:427-434.

[21] 彭张林,张强,杨善林.综合评价理论与方法研究综述[J].中国管理科学,2015,23(S1):245-256.

[22] 张梦中,马克·霍.定性研究方法总论[J].中国行政管理,2011,(11):39-42.

[23] Saaty T I,Tran L T.On the invalidity of fuzzifying numerical judgments in the Analytic Hierarchy Process[J].Mathematical and Computer Modelling,2007,46(7-8):962-975.

[24] 张晓慧,冯英浚.一种非线性模糊综合评价模型[J].系统工程理论与实践,2005,25(10):54-59.

[25] Abdi H, Williams L J. Principal component analysis[J]. Wiley Interdisciplinary Reviews: Computational Statistics,2010,2(4):433-459.

[26] Bojkovi? N, Ani? I, Pej?i?-Tarle S. One solution for cross-country transport-sustainability evaluation using a modified ELECTRE method[J]. Ecological Economics, 2010, 69(5): 1176-1186.

[27] Shyur H J. COTS evaluation using modified TOPSIS and ANP[J]. Applied mathematics and computation, 2006, 177(1): 251-259.

[28] RAZAVIAN A S, AZIZPOUR H, SULLIVAN J，et aLCNN Features Off-the-Shelf: An Astounding Base line for Re-cognition[C]// IEEE Conference on Computer Vision and Pattern Recognition Workshops. IEEE Com-puter Society, 2014 ; 512-519.

[29] AGRAWAL P, GIRSHICK R, MALIK J. Analyzing the Performance of Multilayer Neural Networks for Object Recognition巨J口.Lecture Notes in Computer Science,2014,8695:329-344.

[30] He K, Zhang X, Ren S, et al. Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2014, 37(9):1904-16.

[31] GGRAWAL P, GIRSHICK R, MALIKR. Fast R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision. 2015:1440-1448.REN S, HE K, GIRSHICK R, et al. Faster R-CNN: to-wards real-time object detection with region proposalnetworks// International Conference on Neural In-formation Processing Systems. MIT Press, 2015:91-99.

[32] Ren S, He K, Girshick R, et al. Faster r-cnn: Towards real-time object detection with region proposal networks[C]. Advances in neural information processing systems. 2015: 91-99.

[33] REN S, HE K, GIRSHICK R, et al. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks[J]. arXiv:1506. 01497,2015.

[34] Redmon J,Farhadi A.YOLO9000:Better,Faster,Stronger[C].// Proc of IEEE Conference on Computer Vision and Pattern Recognition,2017.

[35] 徐昆,朱国华,刘文凤,范超.YOLO算法在目标姿态检测中的应用[J].电子技术与软件工程,2019(16):181-182.

[36] 吴雨露,张德贤.基于深度学习的目标检测算法综述[J].信息与电脑(理论版),2019(12):46-48.

[37] Dai J, Qi H, Xiong Y, et al. Deformable convolutional networks[C]//Proceedings of the IEEE international conference on computer vision. 2017: 764-773.

[38] Zhu C, Han S, Mao H, et al. Trained ternary quantization[J]. arXiv preprint arXiv:1612.01064, 2016.

[39] 陈文兵,管正雄,陈允杰.基于生成式对抗神经网络的数据增强方法[J/OL].计算机应用:1-8[2020-03-10].

[40] 周志华.机器学习[M].清华大学出版社,2016.

[41] P. Ekman, W. Friesen, Constants across cultures in the face and emotion. Journal of Personality and Social Psychology, 17(2): 124-129, 1971.

[42] Sara Sabour, Nicholas Frosst, Geoffrey E. Hinton. Dynamic Routing Between Capsules[C].// in 31 st Conference on Neural Information Processing Systems (NIPS 2017), Long Beach, CA, USA.

[43] 付家慧, 吴晓富, 张索非. 基于仿射变换的胶囊网络特征研究[J]. 信号处理, 2018, 034(012):1508-1516.

[44] K. Simonyan, A. Zisserman, Very deep convolutional networks for large-scale image recognition (2014). arXiv preprint arXiv:1409.1556

[45] Chen T , Li M , Li Y , et al. MXNet: A Flexible and Efficient Machine Learning Library for Heterogeneous Distributed Systems[J]. Statistics, 2015: 34-38.