服装图像具有明暗不一、尺度各异的特性,这使得已有识别方法表现不佳。为解决此问题,本文基于空间注意力选择核(space attention selective kernel,SASK)模块和双分支结构搭建神经网络模型对服装图像进行识别。首先,结合跳跃连接、稠密连接和多尺度、通道拆分的思想,设计双分支神经网络,充分提取服装对象的整体特征。其次,基于空间注意力机制,设计SASK模块,使网络可以更多地关注服装对象的形态特征信息,从而提升识别效果。实验结果表明,本文所提方法不但在典型服装数据集上能够取得优于现有主流方法的识别精度,而且在具有明暗不一、尺度各异特性的其他图像数据集上同样表现良好。
Due to the limited performance of the existing recognition methods for clothing images with varying brightness and scales, we propose a neural network model based on spatial attention selective kernel（SASK） and double branch structure in this paper.Firstly, a double branch neural network that incorporates jump connection, dense connection, multi-scale and channel splitting is established to fully extract the overall features of clothing images. Secondly, the SASK module based on the spatial attention mechanism is designed to enable the network to focus on the morphological feature information of clothing images for accurate recognition. Experimental results show that our method not only improves the recognition accuracy of typical clothing datasets compared to existing mainstream methods, but also performs effectively on image datasets with different brightness and scales.
[1] HuAnG Q B, HAn X P, Lu T, et al. Clothing image retrieval based on parts detection and segmentation[C]//20213rd International Conference on Image Processing and Machine Vision(IPMV), 2021:53-59.
[2] 赵宏伟,刘晓涵,张媛,等.基于关键点注意力和通道注意力的服装分类算法[J].吉林大学学报(工学版), 2020, 50(5):1765-1770.ZhAo H W, Liu X H, ZhAnG Y, et al. Clothing classification algorithm based on landmark attention and channel attention[J]. Journal of Jilin University(Engineering and Technology Edition), 2020, 50(5):1765-1770.(in Chinese)
[3] 李巨虎,范睿先,陈志泊.基于颜色和纹理特征的森林火灾图像识别[J].华南理工大学学报(自然科学版), 2020, 48(1):70-83.Li J H, FAn R X, ChEn Z B. Forest fire recognition based on color and texture features[J].Journal of South China University of Technology(Natural Science Edition), 2020, 48(1):70-83.(in Chinese)
[4] 郑长亮,庞明.基于卷积神经网络的时空权重姿态运动特征提取算法[J].应用科学学报, 2021,39(4):594-604.ZhEnG C L, PAnG M. Spatial-temporal weight attitude motion feature extraction algorithm using convolutional neural network[J]. Journal of Applied Sciences, 2021, 39(4):594-604.(in Chinese)
[5] 赵云山,段友祥.基于Attention机制的卷积神经网络文本分类模型[J].应用科学学报, 2019, 37(4):541-550.ZhAo Y S, DuAn Y X. Convolutional neural networks text classification model based on attention mechanism[J]. Journal of Applied Sciences, 2019, 37(4):541-550.(in Chinese)
[6] 彭宁,陈爱斌,周国雄,等.基于正弦注意力表征网络的环境声音识别[J].应用科学学报, 2021,39(4):641-649.PEnG N, ChEn A B, Zhou G X, et al. Environmental sound recognition based on attention sinusoidal representation network[J].Journal of Applied Sciences, 2021, 39(4):641-649.(in Chinese)
[7] 陈巧红,陈翊,李文书,等.多尺度SE-Xception服装图像分类[J].浙江大学学报(工学版), 2020,54(9):1727-1735.ChEn Q H, ChEn Y, Li W S, et al. Clothing image classification based on multi-scale SEXception[J]. Journal of Zhejiang University(Engineering Science), 2020, 54(9):1727-1735.(in Chinese)
[8] ELLEuch M, MEzGhAni A, KhEMAkhEM M, et al. Clothing classification using deep CNN architecture based on transfer learning[C]//International Conference on Hybrid Intelligent Systems. Cham:Springer, 2021:240-248.
[9] Ai X Z, ZhuAnG J W, WAnG Y H, et al. ResCaps:an improved capsule network and its application in ultrasonic image classification of thyroid papillary carcinoma[J]. Complex&Intelligent Systems, 2022, 8(3):1865-1873.
[10] Guo Y C, Du L, ChEn J. Max-margin multi-scale convolutional factor analysis model with application to image classification[J]. Expert Systems with Applications, 2019, 133:21-33.
[11] SEo Y, Shin K S. Hierarchical convolutional neural networks for fashion image classification[J]. Expert Systems with Applications, 2019, 116:328-339.
[12] HuAnG W K, Zhou F B. DA-CapsNet:dual attention mechanism capsule network[J]. Scientific Reports, 2020, 10(1):1-13.
[13] Li X, WAnG W H, Hu X L, et al. Selective kernel networks[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR), 2019:510-519.
[14] ChEn J B, HuAnG R Y, ZhAo K, et al. Multiscale convolutional neural network with feature alignment for bearing fault diagnosis[J]. IEEE Transactions on Instrumentation and Measurement, 2021, 70:1-10.
[15] Li J, Liu Y, ZhAnG Y D, et al. Cascaded attention DenseUNet(CADUNet)for road extraction from very-high-resolution images[J]. ISPRS International Journal of Geo-Information, 2021,10(5):329.
[16] HuAnG G, Liu Z, VAn DER MAATEn L, et al. Densely connected convolutional networks[C]//2017 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR), 2017:2261-2269.
[17] PRAJApATi K, ChuDAsAMA V, PATEL H, et al. Channel split convolutional neural network(ChaSNet)for thermal image super-resolution[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops(CVPRW), 2021:4363-4372.
[18] Li R F, Lu W C, LiAnG H Y, et al. Multiple features with extreme learning machines for clothing image recognition[J]. IEEE Access, 2018, 6:36283-36294.
[19] AL BATAinEh A, KAuR D. Optimal convolutional neural network architecture design using clonal selection algorithm[J]. International Journal of Machine Learning and Computing, 2019,9(6):788-794.
[20] MA B T, Li X, XiA Y, et al. Autonomous deep learning:a genetic DCNN designer for image classification[J]. Neurocomputing, 2020, 379:152-161.
[21] BALDoMinos A, SAEz Y, IsAsi P. Hybridizing evolutionary computation and deep neural networks:an approach to handwriting recognition using committees and transfer learning[J].Complexity, 2019:1-16.
[22] Dos SAnTos M M, DA SiLvA FiLho A G, Dos SAnTos W P. Deep convolutional extreme learning machines:Filters combination and error model validation[J]. Neurocomputing, 2019,329:359-369.
