多阶段特征重分布算法的小样本目标检测  

Multi-Stage Feature Redistribution for Few-Shot Object Detection

作　　者：刘露露 贺占庄[1] 马钟 刘彬[1] 王莉[1] LIU Lulu;HE Zhanzhuang;MA Zhong;LIU Bin;WANG Li(Xi’an Microelectronics Technology Institute,Xi’an 710065)

机构地区：[1]西安微电子技术研究所,西安710065

出　　处：《电子科技大学学报》2023年第1期116-124,共9页Journal of University of Electronic Science and Technology of China

基　　金：国家科技重大专项(2017-V-0014-0066)。

摘　　要：深度神经网络在目标检测任务上需要训练大量的标签数据,然而在许多实际应用场景中标签数据难以获取。针对这一问题,提出了一种面向小样本目标检测的多阶段特征重分布算法(MSFR)。该算法通过对特征向量进行重分布变换,解决了小样本任务下源域数据和目标域数据分布不一致的问题;通过多阶段学习策略将源域知识逐步迁移到小样本目标任务中,进一步提高知识迁移效率。在VOC数据集上的大量实验表明,与现有小样本目标检测算法相比,该算法在不同任务上的精度最高提升了9.06%。该算法在大幅提高小样本目标域类别检测性能的同时,较大限度地保持了对源域类别的检测精度,具有较大的实用价值。Deep neural networks(DNN)in object detecting tasks have witnessed significant progress in the past years.However,it relies on intensive training data with accurate bounding box annotations for a remarkable performance.Once the labelled data are hard to catch,the generalization ability of DNN is far from satisfactory.We propose a few-shot object detecting method based on a multi-stage training strategy within feature redistribution(MSFR).Based on the analyses of the distribution of source domain dataset and target domain dataset in few-shot tasks,a feature redistribution algorithm is proposed to make the feature distribution meet Gaussian distribution or quasi-Gaussian distribution.It solves the inconsistency distribution of the source domain dataset and the target domain dataset.Then,a multi-stage training algorithm is proposed,which improves the efficiency of transferring the source domain knowledge to the target domain task when only a small amount of labeled data for training in each class.Thus,our proposed method significantly improves the detection performance of few-shot target domain categories while maximizing the detection accuracy of the source domain categories.The experimental results on VOC datasets show that the proposed algorithm achieves a precision improvement of up to9.06%on different tasks,compared with existing few-shot object detection approaches.

关 键 词：深度神经网络 特征向量重分布 小样本目标检测 多阶段训练 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]