Minority Clothing Recognition based on Improved DenseNet
DOI:
https://doi.org/10.54097/jceim.v10i2.7877Keywords:
Minority clothing, Image recognition, DenseNet, Attention mechanismAbstract
In response to the difficulty of minority clothing recognition in China, this paper proposes an SE-DenseNet-LR recognition model. This model replaces the ReLU activation function in each dense connection block with the Leaky ReLU function, which has the advantages of faster convergence speed, better generalization ability, and mitigating gradient disappearance in model training. The SE (Squeeze-and-Excitation) attention mechanism is added after each convolutional layer in each dense connection block to obtain more important feature information. The convolutional layers in the connection layers are replaced with dilated convolutions to increase the receptive field. The model also employs learning rate decay and image dataset augmentation strategies to prevent overfitting. Experimental results show that the SE-DenseNet-LR model achieves an accuracy of 84.35% in recognizing 20 categories of minority clothing, which is 2.35%, 2.66%, and 1.88% higher than the recognition accuracies of ResNet18, ResNet34, and DenseNet models, respectively. This model has strong feature extraction ability and robustness, which lays a good foundation for minority clothing recognition.
References
Zhu Hengde,Wang Jian,Wang ShuiHua, et al. An Evolutionary Attention-Based Network for Medical Image Classification[J]. International Journal of Neural Systems, 2023, 33(3).
Zhang Yuhan,Luo Luyang,Dou Qi, et al. Triplet attention and dual-pool contrastive learning for clinic-driven multi-label medical image classification[J]. Medical Image Analysis, 2023, 86.
Kolukisa Burak,Yildirim Veli Can,Elmas Bahadir, et al. Deep learning approaches for vehicle type classification with 3-D magnetic sensor[J]. Computer Networks, 2022, 217.
Liu Yuanyuan,Peng Jiyao,Dai Wei, et al. Joint spatial and scale attention network for multi-view facial expression recognition[J]. Pattern Recognition, 2023, 139.
Alex Krizhevsky,Ilya Sutskever,Geoffrey E. Hinton. ImageNet classification with deep convolutional neural networks[J]. Communications of the Acm, 2017, 60(6).
Kaiming He,Xiangyu Zhang,Shaoqing Ren, et al. Deep Residual Learning for Image Recognition.[J]. Corr, 2015, abs/1512.03385.
Karen Simonyan,Andrew Zisserman. Very Deep Convolutional Networks for Large-Scale Image Recognition.[J]. Corr, 2014, abs/1409.1556.
Yang B, Xu D, Zhang H, et al. Recognition of ethnic minority clothing based on improved DenseNet-BC[J]. Journal of Zhejiang University (Science Edition), 2021, 48(6): 676-683.
Hou Hongtao, Wang Wei, Shen Hongting, et al. Research on Image Classification of Ethnic Minority Clothing Based on Adaptive Image Enhancement and CNN. Modern Computer, 2022, 28(24): 29-35.
Zhang Yue, He Xiyue, Zhao Chenglong. Design and Implementation of Ethnic Clothing Recognition System Based on SE-ResNet. Electronic Technology and Software Engineering, 2022, (8): 205-208.
Zeng Fuliang, Hu Wenjin, He Guoyuan, et al. Tangka Image Classification Based on DenseNet. Modern Electronics Technique, 2022, 45(6): 153-157.
He Q, Guan M, Gan L. Improved clothing image segmentation algorithm based on DeepLabv3+. Fujian Computer, 2023, 39(2): 21-26.
Gao Huang,Zhuang Liu 0003,Kilian Q. Weinberger. Densely Connected Convolutional Networks.[J]. Corr, 2016, abs/1608.06993.
Hu Jie,Shen Li,Albanie Samuel, et al. Squeeze-and-Excitation Networks[J]. Ieee Transactions on Pattern Analysis and Machine Intelligence, 2019, 42(8).
Downloads
Published
Issue
Section
How to Cite
