复杂背景下对空红外弱小目标检测方法

Research on Infrared-Image-Based Detection of Dimand Small Targets in Complex Background

红外成像系统具有探测距离远、抗干扰能力强的特点,在武器系统中备受关注。针对当前目标检测算法在复杂背景下准确率低,误检、漏检率高的问题,提出基于ResNet101主干网的Faster RCNN改进网络。通过将多尺度特征提取和多层特征融合,精确获取目标候选区域,提升小目标检测性能;利用残差网络优化模型结构;基于候选框的检测方法,充分考虑不同尺度的区域特征;基于迁移学习的方法解决小样本数据集泛化性差的问题。实验结果表明,所述方法相较于5种代表性方法,具有准确率高、鲁棒性强的特点。

Infrared-Image-Based system has the characteristics of long detection distance and strong anti-interference ability withmuch attention in weapon systems.Aiming at the problems of low accuracy,high false detection and high missed detection rate of the current target detection algorithm in complex background,this paper proposes an improved Faster-RCNN based on ResNet101 backbone network is proposed.Through multi-scale feature extraction and multi-layer feature fusion,the target candidate region is accurately obtained to improve the performance of small target detection.The residual network is used to optimize model structure.The regional characteristics of different scales are considered fully based on the candidate frame detection method.The method based on transfer learning solves the problem of poor generalization of small sample datasets.The experimental results show that compared with five kinds of representative methods,the proposed method has the characteristics of high accuracy and strong robustness.