Application of a new deep learning method with CBAM in clothing image classification
Sheng Yu, Shangzhu Jin, Jun Peng, Haiyang Liu, Yuanyuan He · 2021 IEEE International Conference on Emergency Science and Information Technology (ICESIT) · 2021
How to effectively identify the clothing category from the massive clothing image is one of the hot research in the field of image recognition in recent years. However, many existing recognition methods lack attention to clothing details, resulting in poor generalization and long calculation cycle, which is not suitable for practical application. In this paper, a new network structure called VCG is proposed. Firstly, VGG16 is used as feature extraction network. Then the convolutional block attention module (CBAM) is added to the second convolution block to emphasize the channel and spatial information, enhance the attention to the target area and extract the detail information from the image. Finally, the global average pooling is used to replace the full connection layer to vectorize the features, which greatly reduces the calculation parameters and gives each channel the actual category significance to output the recognized clothing category. The experimental results show that compared with several common neural network models, VCG is more accurate and has a good detection effect on clothing category recognition.