自动化学报2025，Vol.51Issue(7)：1525-1545,21.DOI:10.16383/j.aas.c240341
基于最小化背景判别性知识的小样本目标检测算法
Minimizing Background Discriminative Knowledge for Few-shot Object Detection
摘要
Abstract
In the field of few-shot object detection,the"training and fine-tuning"two-stage representation learning framework is widely used due to the simplicity of its learning strategy.However,through exploratory experiments,we demonstrate that this learning paradigm is prone to misclassify novel instances as background instances,which hinders the ability of the model to recognize novel object instances.To address this issue,we propose that con-struct a regularized classifier and use the background discriminative knowledge minimizing regulator(BDKMR)to guide the classifier training.BDKMR explicitly reduces the effect of background discriminative knowledge on the classifier for novel categories by employing the background discriminative knowledge minimizing cross-lp regulariza-tion.Moreover,BDKMR uses the weight norm manager to adjust the weight norm of each category in the classifier in order to enhance the model's attention to new categories,while alleviating its bias toward the background cat-egory.Additionally,considering that BDKMR can alter the feature space distribution,the decoupled box classifier module is introduced to adjust the impact of the regulator on the feature extractor during the fine-tuning stage.Ex-perimental results on multiple datasets validate that the proposed method effectively reduces the misclassification of novel object instances and improves the performance of novel categories.关键词
小样本学习/小样本目标检测/正则项/判别性知识Key words
Few-shot learning/few-shot object detection/regularization/discriminative knowledge引用本文复制引用
张雅楠,宋飞,靳毅凡,王晓明,刘立祥,李江梦..基于最小化背景判别性知识的小样本目标检测算法[J].自动化学报,2025,51(7):1525-1545,21.基金项目
国家基础科研计划(JCKY2022130C020)资助Supported by Fundamental Research Program,China(JCKY2022130C020) （JCKY2022130C020）
