火力与指挥控制2023，Vol.48Issue(12)：142-147,6.DOI:10.3969/j.issn.1002-0640.2023.12.021
基于可变卷积与迁移学习的小样本检测方法
Few Shot Detection Method Based on Deformable Convolution and Transfer Learning
宋程程 1李捷 1高晓利 1王维 1赵火军1
作者信息
- 1. 四川九洲电器集团有限责任公司,四川 绵阳 621000
- 折叠
摘要
Abstract
Few shot detection method based on deformable convolution and transfer learning is proposed to solve the problem of low efficiency of weak and small target recognition by the mainstream methods.Firstly,based on the idea of deformable convolution,the feature extraction backbone network is improved to achieve the equivalent learning ability on a small amount of data to the ordinary convolution on a large amount of data.Secondly,a soft-NMS interaction action method is designed to reduce the missed detection problem of multi-target overlapping.The experiments on the public data set Pascal VOC and the measured weak and small target data set show that the performance of the improved algorithm is significantly improved compared with the original method,which is 5.5%higher than that of the original algorithm on the public data set and 8.3%higher than that of the original algorithm on the measured data set under the few shot conditions.关键词
小样本检测/弱小目标/可变卷积/迁移学习/特征提取Key words
few shot detection/weak and small targets/deformable convolution/transfer learning/feature extraction分类
信息技术与安全科学引用本文复制引用
宋程程,李捷,高晓利,王维,赵火军..基于可变卷积与迁移学习的小样本检测方法[J].火力与指挥控制,2023,48(12):142-147,6.
