[1]毕忠勤,单美静,刘志斌,等.基于改进 SSD 的少样本目标检测[J].计算机技术与发展,2023,33(11):35-40.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 006]
　BI Zhong-qin,SHAN Mei-jing,LIU Zhi-bin,et al.Few-shot Object Detection Based on Improved SSD[J].,2023,33(11):35-40.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 006]

点击复制

基于改进 SSD 的少样本目标检测()

分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

33

期数:

2023年11期

页码:

35-40

栏目:

媒体计算

出版日期:

2023-11-10

文章信息/Info

Title:

Few-shot Object Detection Based on Improved SSD

文章编号:

1673-629X(2023)11-0035-06

作者:

毕忠勤¹ ; 单美静² ; 刘志斌¹ ; 徐富强¹

1. 上海电力大学 计算机科学与技术学院, 上海 200090;

2. 华东政法大学 信息科学与技术系,上海 201620

Author(s):

BI Zhong-qin1 ; SHAN Mei-jing2 ; LIU Zhi-bin1 ; XU Fu-qiang1

1. School of Computer Science and Technology,Shanghai University of Electric Power,Shanghai 200090,China;

2. Department of Information Science and Technology,East China University of Political Science and Law,Shanghai 201620,China

关键词:

目标检测; 机器学习; 少样本; FPN; SSD 网络模型

Keywords:

object detection; machine learning; few-shot; FPN; SSD

分类号:

TP393

DOI:

10. 3969 / j. issn. 1673-629X. 2023. 11. 006

摘要:

目标检测作为深度学习的热点问题之一,在自动驾驶、行人识别、智能医疗、机器人视觉等多个领域有着广泛的应用前景。 但现有的大部分目标检测模型都依赖于大规模的标注数

据集来训练模型以保证目标检测的准确率,而在许多实际的应用场景中,大量数据的标注不仅耗费人力物力,而且需要大量专业人士的参与,在一定程度上限制了目标检测模型的实际应用。 针对少样本目标检测的特殊要求,基于 SSD 网络提出了一种改进的少样本目标检测模型,提高了目标检测应用的适用性。 首先,在 SSD ( Single Shot multiBox Detector) 网络的基础上,用 ResNet-50 代替 VGG 作为特征网络,从而提升模型的特征提取能力。 其次,通过引入残差单元避免了网络退化问题。 最后,为了充分融合各层之间的语义信息和位置信息,用 FPN ( Feature Pyramid Networks)替换了原模型中间的两个特征层。 基于改进 SSD 网络的目标检测模型在少样本数据集的检测结果中,mAP 值达到了 79.8% ,比原始模型提高了 2. 6 百分点。

Abstract:

As one of the hot topics of deep learning,object detection has a wide application prospect in many fields,such as automaticdriving,pedestrian recognition,intelligent medical treatment,and robot vision and so on. However,most of the existing object detectionmodels rely on large-scale annotation data sets for model training to ensure the accuracy?

of target detection. In many practical applicationscenarios,the annotation of a large number of data not only consumes human and material resources,but also requires the participation ofa large number of professionals,which limits the practical application of the object detection model to a certain extent. Aiming at thespecial requirements of few-

shot object detection,we propose an improved few-shot object detection model based on SSD ( Single ShotmultiBox Detector) network,which improves the applicability of object detection applications. Firstly,the ResNet-50 is used instead ofVGG as the feature network on the basis of SSD network to improve the feature extraction capability of the model. Secondly,the problemof network degradation is avoided by introducing residual element. Finally, in order to fully integrate the semantic information andlocation information between the layers,FPN is used to replace the two feature layers in the middle of the original model. In the detectionresults of the target detection model based on the improved SSD network in a few-shot data set,the mAP value of the improved modelreached 79. 8% ,which was 2. 6 percentage points higher than that of the original model.

常用功能