| 摘要: |
| 目的 为快速准确识别消费者对产品意象的认知，提出一种基于分类器链的产品意象识别方法。方法 首先，构建产品意象数据集，通过相似性聚类和网络爬虫得到产品意象词与产品图像，在此基础上，进行产品意象实验，获得消费者对于产品意象的认知，构建产品意象数据集;然后，提取图像特征，利用卷积神经网络RestNet50提取产品图像特征;最后，使用分类器链算法构建产品意象识别模型，提出基于混淆矩阵与条件熵的分类器链标签顺序确定方法，确定产品意象标签顺序。结论 为了验证所述标签顺序确定方法在识别产品意象中具有优越性设计了对比实验。实验结果表明，相较于其他方法，基于分类器链的产品多标签意象识别方法考虑了标签的识别结果与相关关系，能显著提升模型对于产品多标签意象的预测性能。 |
| 关键词: 产品意象 多标签分类 分类器链 深度学习 |
| DOI：10.19554/j.cnki.1001-3563.2021.14.004 |
| 分类号:TB472 |
| 基金项目:国家重点研发计划资助项目(2019YFB1405701) |
|
| Product Multi-label Image Recognition Method Based on Classifier Chain |
|
CHU Jian-jie, WANG Peng-chao, CHEN Chen, SHI Ying-xi
|
|
(Key Laboratory of Industrial Design and Ergonomics, Ministry of Industry and Information Technology, Northwestern Polytechnical University, Xi’an 710072, China)
|
| Abstract: |
| In order to identify consumers’ cognition of product image in a quick and accurate manner, a product image recognition algorithm based on classifier chain is proposed. First, product image data sets are built, and product labels and pictures are obtained by similarity clustering and web crawler. On this basis, consumers' cognition of product image is obtained and product image data sets are built through product image experiment. The next step is to extract the image features. The product image features are extracted using the convolutional neural network RESTNet50, then the product image recognition model is constructed using the classifier chain algorithm, and, based on confusion matrix and conditional entropy, the label order of product image is determined. A comparative experiment is designed to verify that the algorithm has advantages in product image recognition. The experimental results show that, compared with other methods, it takes into account the label identification results and the related relationship, thus can significantly improve prediction performance of the model for product multi-label image. |
| Key words: product image multi-label classification classifier chain deep learning |
渝公网安备 50010702501716号