图双线性池化特征编码的细粒度目标识别方法
Fine-Grained Object Recognition Method Based on Graph Bilinear Pooling Feature Encoding
芮挺 1杜晓明 2王东 1郑南 1史建军1
作者信息
- 1. 陆军工程大学野战工程学院,江苏南京 210007
- 2. 陆军工程大学指挥控制工程学院,江苏南京 210007
- 折叠
摘要
针对细粒度图像识别领域中经典双线性池化模型存在的视觉突发与特征冗余问题,提出了 一种图双线性池化模型.该模型将图网络嵌入双线性池化模型,利用图网络的聚合能力,将差异性图像特征编码为高阶特征,改善了编码过程中的视觉突发现象.在CUB、Cars和Aircrafts 3个公共数据集上进行实验,模型的精确度分别达到87.8％、93.5％和89.6％.相较于分解双线性池化,该模型参数量仅为基线模型的25％,识别精度分别提高2.4％、1.7％和1.3％,充分验证了模型的有效性,可为军事目标细粒度识别提供方法参考.
Abstract
To address the problems of visual burst and feature redundancy in the classical bilinear poo-ling model in the field of fine-grained image recognition,this paper proposes a graph bilinear pooling mod-el.This model integrates graph networks into the bilinear pooling framework,leveraging the aggregation capabilities of graph networks to encode differential image features into higher-order features,thereby alle-viating the phenomenon of visual burst during the encoding process.The results of the experiments con-ducted on the three public datasets of CUB,Cars and Aircrafts show that the proposed model achieves ac-curacies of 87.8％,93.5％and 89.6％,respectively.Compared with decomposed bilinear pooling,this model's parameter count is only 25％of the baseline model,while the recognition accuracy is improved by 2.4％,1.7％,and 1.3％,respectively,which fully verifies the effectiveness of the model and can provide a method reference for fine-grained recognition of military targets.
关键词
细粒度识别/高阶特征编码/双线性池化/图神经网络Key words
fine-grained recognition/higher-order feature encoding/bilinear pooling(BP)/graph neural network引用本文复制引用
出版年
2024
NETL
NSTL
万方数据