深度度量学习的多视角高频工件图像检索

Multi-View High-Frequency Workpiece Image Retrieval Based on Deep Metric Learning

针对工业场景下高频工件多视角识别精度低的问题,提出一种深度度量学习的高频工件图像检索方法。首先搭建基于卷积神经网络的图像特征向量编码模型,采用差异性损失(Different Loss)提取工件图像的私有特征和公共特征,通过相似度损失(Similarity Loss)融合多视角图像的公共特征获得初级嵌入向量;然后利用三元组中心损失(Triplet-Center Loss)以减小类内距离和增大类间距离为准则监督嵌入向量的学习,获得鲁棒性强的嵌入向量;最后以该嵌入向量表示高频工件图像的特征编码,实现多视角高频工件的图像检索。实验结果表明,提出的方法比单视角特征编码具有更强的表征能力,其检索准确率提高了8.95%;在相同网络结构下,提出的模型比其他网络模型的检索准确率高出9.68%。

Aiming at the problem of low accuracy of multi-view recognition of high-frequency workpieces in industrial scenes,a high-frequency workpiece image retrieval method based on deep metric learning is proposed.First,build an image feature vector encoding model based on convolutional neural network,use Different Loss to extract the private and public features of the artifact image,and use the Similarity Loss to fuse the public features of the multi-view image to obtain the primary embedding Vector;then use Triplet-Center Loss to supervise the learning of the embedding vector based on the criteria of reducing the distance within the class and increasing the distance between classes,and obtain a robust embedding vector;finally represented by the embedding vector the feature coding of high-frequency workpiece images realizes image retrieval of high-frequency workpieces from multiple perspectives.Experimental results show that the proposed method has stronger characterization ability than single-view feature coding,and its retrieval accuracy is increased by 8.95%;under the same network structure,the retrieval accuracy of the proposed model is 9.68%higher than that of other network models.