| 摘要: |
| 在深度学习领域中，注意力机制因其出色的性能以及即插即用的便捷性，在图像处理任务中得到了广泛应用。介绍了通道注意力机制、空间注意力机制以及通道与空间混合注意力机制这3类主流注意力机制的核心思想和实现方法。通过对比分析它们之间的优势与缺陷，探讨了注意力机制所存在的挑战与问题,给出了采用VGGNet（Visual Geometry Group Network）模型对注意力机制在图像分类任务中的性能评测结果。最后，展望了注意力机制未来的发展趋势，以期为后续研究提供有价值的参考与启示。 |
| 关键词: 图像处理 图像分类 深度学习 注意力机制 |
| DOI：10.20079/j.issn.1001-893x.241105001 |
|
| 基金项目:四川省科技计划项目(2024YFHZ0026) |
|
| A Survey of Attention Mechanisms in Image Classification Tasks |
| LIN Guojuna,b,GONG Taoa |
| (1a.Sichuan Provincial Key Laboratory of Artificial Intelligence;1b.College of Automation and Information Engineering,Sichuan Light Chemical Engineering University,Yibin 644000,China;2.School of Intelligence,Peking University,Beijing 100871,China) |
| Abstract: |
| In the field of deep learning,the attention mechanism，due to its outstanding performance and the convenience of plug-and-play,has been widely applied in image processing tasks.The authors introduce the core ideas and implementation methods of three mainstream attention mechanisms including channel attention mechanism,spatial attention mechanism,and channel and spatial mixed attention mechanism,discuss the challenges and problems existing in the attention mechanism by comparing and analyzing their advantages and disadvantages,provide the performance evaluation result of visual geometry group network(VGGNet) model in image classification tasks,and finally look forward to the future development trends of the attention mechanism,in hope of providing valuable references and inspirations for subsequent research. |
| Key words: image processing image classification deep learning attention mechanism |
《电讯技术》编辑部 