西装识别的深度学习方法
|
摘要
为准确而快速地对电商平台产品图像进行西装目标的分类检测,以3个主要的卷积网络深度学习框架即快速区域卷积神经网络、基于区域的全连接卷积网络和单次多盒检测为基础,首先通过实验分析其在服装图像分类识别中的效率和有效性,针对小目标识别困难和过拟合识别问题,提出基于尺寸分割和负样本的单次多盒检测(SSD)增强方法(DN-SSD);然后将图像分割为不同尺寸的子图突出服装目标,通过融合分类方法解决SSD算法对小目标识别不足的问题,并通过增强负样本以提高算法的场景适应能力。实验结果表明,该算法可有效地识别各种形态和大小的西装目标,识别准确率达到90%以上,并且能够方便地推广到服装其他品类的识别中。
In order to classify and detect the suit target in images of e-commerce platform accurately and quickly,an enhanced deep convolution network( DN-SSD) was proposed. First,three main frameworks faster region-convolutional networks( faster R-CNN),region-based fully convolution network( R-FCN)and single shot muti-box detection( SSD) were evaluated. An image was segmented into multiscale subimages to highlight the suit target based on the SSD. Secondly,the problem of small target recognition was solved by the fusion of classification. The scene adaptability was enhanced by increasing number of negative samples. The experimental result shows that the algorithm can recognize various shapes and size of suit targets and achieves the accuracy over 90%. The method can also be generalized to other style of dress detection and location.
In order to classify and detect the suit target in images of e-commerce platform accurately and quickly,an enhanced deep convolution network( DN-SSD) was proposed. First,three main frameworks faster region-convolutional networks( faster R-CNN),region-based fully convolution network( R-FCN)and single shot muti-box detection( SSD) were evaluated. An image was segmented into multiscale subimages to highlight the suit target based on the SSD. Secondly,the problem of small target recognition was solved by the fusion of classification. The scene adaptability was enhanced by increasing number of negative samples. The experimental result shows that the algorithm can recognize various shapes and size of suit targets and achieves the accuracy over 90%. The method can also be generalized to other style of dress detection and location.
引文
[1]纪娟,秦珂,杨若瑜.基于HOG和几何特征的服装细节要素识别与分类[J].图学学报,2016,37(1):84-90.JI Juan,QIN Ke,YANG Ruoyu. Classification of the detail features for clothes based on HOG and geometric features[J]. Journal of Graphics,2016,37(1):84-90.
[2]魏芬,刘建平,徐松松,等.基于多特征值的服装检测与识别算法[J].实验室研究与探索,2016,35(5):118-122.WEI Fen,LIU Jianping,XU Songsong,et al. Research on clothing detection and recognition algorithm based on characteristic values[J]. Research and Exploration in Laboratory,2016,35(5):118-122.
[3]李东,万贤福,汪军.采用傅里叶描述子和支持向量机的服装款式识别方法[J].纺织学报,2017,38(5):122-127.LI Dong, WAN Xianfu, WANG Jun. Clothing style recognition approach using Fourier descriptors and support vector machines[J]. Journal of Textile Research,2017,38(5):122-127.
[4]李东,万贤福,汪军,等.基于轮廓曲率特征点的服装款式识别方法[J].东华大学学报(自然科学版),2018(1):1-6.LI Dong,WAN Xianfu,WANG Jun,et al. Clothing style recognition approach based on the curvature feature points on the contour[J]. Journal of Donghua University(Natural Science Edition),2018(1):1-6.
[5]彭刚,杨诗琪,黄心汉,等.改进的基于区域卷积神经网络的微操作系统目标检测方法[J].模式识别与人工智能,2018,31(2):142-149.PENG Gang,YANG Shiqi,HUANG Xinhan,et al.Improved object detection method of micro-operating system based on region convolutional neural network[J].Pattern Recognition and Artificial Intelligence,2018,31(2):142-149.
[6] KRIZHENVSHKY A, SUTSKEVER I, HINTON G.Imagenet classification with deep convolutional networks[C]//Proceedings of the Conference Neural Information Processing Systems(NIPS).[S.l.]:Curran Associates Int,2012:1097-1105.
[7] PANG Y,SUN M,JIANG X,et al. Convolution in convolution for network in network[J]. IEEE Transactions on Neural Networks&Learning Systems,2018,29(5):1587-1597.
[8] NODA K,YAMAGUCHI Y,NAKADAI K,et al. Audio-visual speech recognition using deep learning[J].Applied Intelligence,2015,42(4):722-737.
[9] ZHANG K, SUN M, HAN T X, et al. Residual networks of residual networks:multilevel residual networks[J]. IEEE Transactions on Circuits&Systems for Video Technology,2016(99):1.
[10]范荣.基于卷积神经网络的服装种类识别[J].现代计算机,2016(9):29-32.FAN Rong. Classification of clothing type based on convolutional neural network[J]. Modern Computer,2016(9):29-32.
[11] REN S,HE K,GIRSHICK R,et al. Faster R-CNN:towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Mach Intell,2016,39(6):1137-1149.
[12] HE K M,GKIOXARI G,DOLLAR P,et al. Mask RCNN[C]//Proceedings of the IEEE International Conference on Computer Vision(ICCV). Venice:IEEE,2017:2980-2988.
[13] LIU W,ANGUELOV D,ERHAN D,et al. SSD:single shot multibox detector[C]//Proceedings of European Conference on Computer Vision. Amsterdam:ECCV,2016:21-37.
[14]唐聪,凌永顺,郑科栋,等.基于深度学习的多视窗SSD目标检测方法[J].红外与激光工程,2018,47(1):290-298.TANG Cong,LING Yongshun,ZHENG Kedong,et al.Object detection method of multi-view SSD based on deep learning[J]. Infrared and Laser Engineering,2018,47(1):290-298.
[2]魏芬,刘建平,徐松松,等.基于多特征值的服装检测与识别算法[J].实验室研究与探索,2016,35(5):118-122.WEI Fen,LIU Jianping,XU Songsong,et al. Research on clothing detection and recognition algorithm based on characteristic values[J]. Research and Exploration in Laboratory,2016,35(5):118-122.
[3]李东,万贤福,汪军.采用傅里叶描述子和支持向量机的服装款式识别方法[J].纺织学报,2017,38(5):122-127.LI Dong, WAN Xianfu, WANG Jun. Clothing style recognition approach using Fourier descriptors and support vector machines[J]. Journal of Textile Research,2017,38(5):122-127.
[4]李东,万贤福,汪军,等.基于轮廓曲率特征点的服装款式识别方法[J].东华大学学报(自然科学版),2018(1):1-6.LI Dong,WAN Xianfu,WANG Jun,et al. Clothing style recognition approach based on the curvature feature points on the contour[J]. Journal of Donghua University(Natural Science Edition),2018(1):1-6.
[5]彭刚,杨诗琪,黄心汉,等.改进的基于区域卷积神经网络的微操作系统目标检测方法[J].模式识别与人工智能,2018,31(2):142-149.PENG Gang,YANG Shiqi,HUANG Xinhan,et al.Improved object detection method of micro-operating system based on region convolutional neural network[J].Pattern Recognition and Artificial Intelligence,2018,31(2):142-149.
[6] KRIZHENVSHKY A, SUTSKEVER I, HINTON G.Imagenet classification with deep convolutional networks[C]//Proceedings of the Conference Neural Information Processing Systems(NIPS).[S.l.]:Curran Associates Int,2012:1097-1105.
[7] PANG Y,SUN M,JIANG X,et al. Convolution in convolution for network in network[J]. IEEE Transactions on Neural Networks&Learning Systems,2018,29(5):1587-1597.
[8] NODA K,YAMAGUCHI Y,NAKADAI K,et al. Audio-visual speech recognition using deep learning[J].Applied Intelligence,2015,42(4):722-737.
[9] ZHANG K, SUN M, HAN T X, et al. Residual networks of residual networks:multilevel residual networks[J]. IEEE Transactions on Circuits&Systems for Video Technology,2016(99):1.
[10]范荣.基于卷积神经网络的服装种类识别[J].现代计算机,2016(9):29-32.FAN Rong. Classification of clothing type based on convolutional neural network[J]. Modern Computer,2016(9):29-32.
[11] REN S,HE K,GIRSHICK R,et al. Faster R-CNN:towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Mach Intell,2016,39(6):1137-1149.
[12] HE K M,GKIOXARI G,DOLLAR P,et al. Mask RCNN[C]//Proceedings of the IEEE International Conference on Computer Vision(ICCV). Venice:IEEE,2017:2980-2988.
[13] LIU W,ANGUELOV D,ERHAN D,et al. SSD:single shot multibox detector[C]//Proceedings of European Conference on Computer Vision. Amsterdam:ECCV,2016:21-37.
[14]唐聪,凌永顺,郑科栋,等.基于深度学习的多视窗SSD目标检测方法[J].红外与激光工程,2018,47(1):290-298.TANG Cong,LING Yongshun,ZHENG Kedong,et al.Object detection method of multi-view SSD based on deep learning[J]. Infrared and Laser Engineering,2018,47(1):290-298.