针对传统人体部位体型分类方法费时费力、成本较高的问题，设计一种融合注意力机制的体型分类网络(Attention Body Classification Net，A_BCN)。该网络由弱监督的注意力学习和数据增强两个模块组成，其中:弱监督的注意力学习模块通过注意力机制获得注意力图;数据增强模块通过注意力图指导图像的数据增强，包括注意力裁剪、注意力丢弃和注意力平均。将增强后的图像重新输入到网络中得到特征图，将得到的特征图和注意力图融合进行分类。在后续自制的人体图像数据集中，该算法准确率为90。52％，提高了分类准确率并节省了成本。
FINE-GRAINED CLASSIFICATION METHOD OF HUMAN BODY PARTS INCORPORATING ATTENTION MECHANISM
Aiming at the problem of time-consuming,laborious and costly classification of traditional body parts and body types,this paper designs a body type classification network(A_BCN)that incorporates attention mechanism fine-grained classification.The network consisted of two modules:weakly supervised attention learning and data enhancement.The weakly supervised attention learning module obtained the attention map of the human body through the attention mechanism,and the data enhancement module guided the data enhancement of the image through the attention map,including attention cropping,attention dropping and attention averaging.The enhanced image was re-input into the network to get the feature map,and the feature map and attention map were fused for classification.In the subsequent self-made human body image data set,the accuracy of the algorithm is 90.52％,which improves the classification accuracy and saves the cost.
Body partsBody shape classificationAttention mechanismFine-grained classificationData enhancement
万方数据
维普
