随着我国经济的飞速发展和交通等基础设施的进一步完善，各个城市的汽车数量也在快速增长，如何对大量的汽车和复杂的交通情况进行有效的管理已成为管理部门面临的主要挑战。车型识别应用在很多领域，比如智能交通，无人驾驶以及商业应用等。由于车辆种类品牌型号众多，不同型号车辆间差异较小，所以对车型的细粒度识别是十分必要的，其在智慧交通，追踪车辆，拍照识车等方面都有广泛的应用。目前车型的分类方法主要有传统方法与基于深度学习的方法。传统的图像识别方法通常通过人工对图像进行特征提取，再通过分类器进行分类。而随着深度学习技术日益更新，卷积神经网络作为其重要的分支，在图像识别领域得到了非常广泛的应用，也取得了较好的效果，但在细粒度识别方面的应用还有待深入研究。因此，本文研究基于深度卷积神经网络对细粒度车型识别，以提高识别的准确率。

    本文工作主要体现在以下两个方面：第一，针对普通神经网络特征提取较为单一和网络退化的问题，提出一种基于Inc-Resnet的细粒度车型识别方法。通过采用残差结构来解决随着网络层数加深网络退化的问题，并在其中引入Inception模块，通过并列的不同大小的卷积核分别对车型图像的不同特征进行提取，最后再进行聚合，通过不同大小的感受野来获取不同尺度的特征，进一步提高识别准确率。第二，针对细粒度图像不同类别间差异较小、不同特征间距离较近难以准确识别的问题，提出基于ISC-Resnet的车型识别方法。对损失函数部分进行优化，将softmax交叉熵损失函数和中心损失函数相结合，联合对网络模型进行监督，以提高类内的紧凑性，进而提高类间的分散性，对不同类别进行有效区分，从而在整体上达到提高识别准确率的效果。

本文实验采用斯坦福大学车型数据库stanford cars作为数据源，测试集识别准确率为92.3%。实验结果表明，本文方法能够有效地提取车型特征，提升网络的识别准确率。

With the rapid development of our country's economy and the further improvement of transportation and other infrastructure, the number of cars in each city is also growing rapidly. How to effectively manage a large number of cars and complex traffic conditions has become a major challenge for the management department. Vehicle recognition is applied in many fields, such as intelligent transportation, unmanned driving, and commercial applications. Due to the large number of vehicle types, brands and models, and the small differences between different types of vehicles, fine-grained recognition of vehicle models is very necessary. It has a wide range of applications in smart transportation, tracking vehicles, and taking pictures to identify vehicles. At present, the classification methods of vehicle mainly include traditional methods and methods based on deep learning. Traditional image recognition methods usually extract features from images manually, and then classify them by classifiers. With the increasing update of deep learning technology, convolutional neural network, as an important branch, has been widely used in the field of image recognition and has achieved good results. However, the application of fine-grained recognition still needs to be studied in depth. . Therefore, this paper studies fine-grained car recognition based on deep convolutional neural networks to improve the accuracy of recognition.

The work of this paper is mainly embodied in the following two aspects: First, in view of the relatively single feature extraction of ordinary neural networks and the problem of network degradation, a fine-grained vehicle recognition method based on Inc-Resnet is proposed. The residual structure is used to solve the problem of network degradation as the number of network layers deepens, and the Inception module is introduced into it. Different features of the car image are extracted through parallel convolution kernels of different sizes, and finally aggregated. Different sizes of receptive fields are used to obtain features of different scales, which further improves the recognition accuracy. Second, to solve the problem that the difference between different categories of fine-grained images is small, and the distance between different features is relatively close, it is difficult to accurately identify the problem, and a vehicle recognition method based on ISC-Resnet is proposed. Optimize the loss function part, combine the softmax cross-entropy loss function and the central loss function, and jointly supervise the network model to improve the compactness within the class, thereby improving the dispersion between classes, and effectively distinguishing different categories. So as to achieve the effect of improving the recognition accuracy rate as a whole.

This experiment uses Stanford cars database stanford cars as the data source, and the test set recognition accuracy rate is 92.3%. The experimental results show that the method in this paper can effectively extract vehicle features and improve the recognition accuracy of the network.

[1]张鹏,陈湘军,阮雅端,陈启美.采用稀疏SIFT特征的车型识别方法[J].西安交通大学学报,2015,49(12):137-143.

[2]耿庆田,赵浩宇,于繁华,王宇婷,赵宏伟.基于改进HOG特征提取的车型识别算法[J].中国光学,2018,11(02):174-181.

[3]周书仁,殷建平.基于Haar特性的LBP纹理特征[J].软件学报,2013,24(08):1909-1926.

[4]陈胜,李勃,吕凤花.基于SVM的车型分类方法研究[J].贵州大学学报(自然科学版),2012,29(01):79-82.

[5]张明恒,王华莹,郭烈.基于改进K-Means算法的车辆识别方法[J].计算机技术与发展,2012,22(05):53-56.

[6]Shaoyong Yu,Yun Wu,Wei Li,Zhijun Song,Wenhua Zeng. A model for fine-grained vehicle classification based on deep learning[J]. Neurocomputing,2017,257:97-103.

[7]华莉琴,许维,王拓,马瑞芳,胥博.采用改进的尺度不变特征转换及多视角模型对车型识别[J].西安交通大学学报,2013,47(04):92-99.

[8]LeCun Y, Bottou L, Bengio Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324.

[9]Hinton G E, Salakhutdinov R R. Reducing the dimensionality of data with neural networks[J]. science, 2006, 313(5786): 504-507.

[10]Krizhevsky A, Sutskever I, Hinton G E. Imagenet classification with deep convolutional neural networks[J]. Advances in neural information processing systems, 2012, 25: 1097-1105.

[11]陈立潮,卜楠,潘理虎,曹建芳,张英俊.基于改进卷积神经网络的车型识别[J].计算机工程与设计,2019,40(11):3331-3336+3348.

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

[13]Szegedy C, Liu W, Jia Y, et al. Going deeper with convolutions[C].Proceedings of the IEEE conference on computer vision and pattern recognition. 2015: 1-9.

[14]He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C].Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 770-778.

[15]梁栋,何佳,石陆魁,王松,刘佳.结合运动目标检测和ResNet的车速车型智能识别[J].北京交通大学学报,2019,43(05):10-19.

[16]李晓雨,薛宏飞,高尚,迟万达,范迪.基于改进残差网络的车型识别算法[J].信息技术与信息化,2020(02):121-126.

[17]He K, Zhang X, Ren S, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE transactions on pattern analysis and machine intelligence, 2015, 37(9): 1904-1916.

[18]刘虎,周野,袁家斌.基于多尺度双线性卷积神经网络的多角度下车型精细识别[J].计算机应用,2019,39(08):2402-2407.

[19]Ren S, He K, Girshick R, et al.Faster R-CNN:Towards Real-Time Object Detection with Region Proposal Networks[J].2017,39(6):1137-1149.

[20]曹诗雨,刘跃虎,李辛昭.基于Fast R-CNN的车辆目标检测[J].中国图象图形学报,2017,22(5):0671-0677.

[21]彭秋辰,宋亦旭.基于Mask R-CNN的物体识别和定位[J].清华大学学报（自然科学版）,2019,59(2):135-141.

[22]杨娟,曹浩宇,汪荣贵,薛丽霞,胡敏.区域建议网络的细粒度车型识别［J］.中国图象图形学报, 2018,23(6):0837-0845．

[23]桑军,郭沛,项志立,罗红玲,陈欣.Faster-RCNN的车型识别分析[J].重庆大学学报,2017,40(07):32-36.

[24]Modolo D, Ferrari V. Learning semantic part-based models from google images[J]. IEEE transactions on pattern analysis and machine intelligence, 2017, 40(6): 1502-1509.

[25]Al-Obaydy W N I, Suandi S A. Automatic pose normalization for open-set single-sample face recognition in video surveillance[J]. Multimedia Tools and Applications, 2020, 79(3): 2897-2915.

[26]Wei X S, Xie C W, Wu J, et al. Mask-CNN: Localizing parts and selecting descriptors for fine-grained bird species categorization[J]. Pattern Recognition, 2018, 76: 704-714.

[27]李红艳,李春庚,安居白,任俊丽.注意力机制改进卷积神经网络的遥感图像目标检测[J].中国图象图形学报,2019,24(08):1400-1408.

[28]Goodfellow I J, Pouget-Abadie J, Mirza M, et al. Generative adversarial networks[J]. arXiv preprint arXiv:1406.2661, 2014.

[29]朱秀昌,唐贵进.生成对抗网络图像处理综述[J].南京邮电大学学报(自然科学版),2019,39(03):1-12.

[30]秦兴,宋各方.基于双线性卷积神经网络的猪脸识别算法[J].杭州电子科技大学学报(自然科学版),2019,39(02):12-17.

[31]梁华刚,温晓倩,梁丹丹,李怀德,茹锋.多级卷积特征金字塔的细粒度食物图片识别[J].中国图象图形学报,2019,24(06):870-881.

[32]Dong Z, Wu Y, Pei M, et al. Vehicle type classification using a semisupervised convolutional neural network[J]. IEEE transactions on intelligent transportation systems, 2015, 16(4): 2247-2256.

[33]袁建平,陈晓龙,陈显龙,何恩杰,张加其,高宇豆.基于文本与视觉信息的细粒度图像分类[J].图学学报,2019,40(03):503-512.

[34]宋岩贝,魏维,何冰倩.基于中层特征的细粒度的车型识别[J].计算机工程与设计,2020,41(6):1708-1713.

[35]赵浩如,张永,刘国柱.基于RPN与B- CNN的细粒度图像分类算法研究[J].计算机应用与软件,2019,36(03):210-213+264.

[36]蔺素珍,韩泽.基于深度堆叠卷积神经网络的图像融合[J].计算机学报,2017,40(11):2506-2518.

[37]刘万军,梁雪剑,曲海成.不同池化模型的卷积神经网络学习性能研究[J].中国图像图形学报,2016,21(9):1178-1190.

[38]Xiao T, Xu Y, Yang K, et al. The application of two-level attention models in deep convolutional neural network for fine-grained image classification[C].Proceedings of the IEEE conference on computer vision and pattern recognition. 2015: 842-850.

[39]Dong Z,Pei M,He Y,et al.Vehicle type classification using unsupervised convolutional neural network[C].22nd International Conference on Pattern Recognition,ICPR2014,Stockholm,Swedwn,August,24-28, 2014. Sweden:IEEE, 2014:172-177.

[40]Yang L, Luo P, Change Loy C, et al. A large-scale car dataset for fine-grained categorization and verification[C].Proceedings of the IEEE conference on computer vision and pattern recognition. 2015: 3973-3981.

[41]张顺,龚怡宏,王进军.深度卷积神经网络的发展及其在计算机视觉领域的应用[J].计算机学报,2019,42(3):453-482.

[42]罗建豪,吴建鑫.基于深度卷积特征的细粒度图像分类研究综述[J].自动化学报,2017,43(8):1306-1318.

[43]朱超平,杨艺.基于YOLO2和ResNet算法的监控视频中的人脸检测与识别[J].重庆理工大学学报,2018,32(8):170-175.

[44]李国瑞,何小海,吴晓红,卿粼波,滕奇志.基于语义信息跨层特征融合的细粒度鸟类识别[J].计算机应用与软件,2020,37(04):132-136+191.

[45]金炎,孙伟,唐慧强,张小瑞,陈胜.基于全局和局部特征融合的车型识别[J].计算机工程与设计,2016,37(04):1051-1055.

[46]Wang X Y,Yang M,Zhu S H，et al.Regionlets for generic object detection［J］.IEEE Transactions on Pattern Analysis and Machine Intelligence,2015,37 (10):2071-2084．

[47]Pandey P,Deepthi A,Mandal B，et al．Food Net: recognizing foods using ensemble of deep networks［J］.IEEE Signal Processing Letters，2017，24(12):1758-1762．

[48]吕国豪,罗四维,黄雅平,蒋欣兰.基于卷积神经网络的正则化方法[J].计算机研究与发展,2014,51(09):1891-1900.

[49]Fatemeh Zamani, Mansour Jamzad.A feature fusion based localized multiple kernel learning system for real world image classification[J].EURASIP Journal on Image and Video Processing,2017,78:1-15.

[50]BAOQI LI, AND YUYAO HE.An Improved ResNet Based on the Adjustable Shortcut Connections[J].IEEE Access,2018,18967-18974.