基于目标特征分布增强卷积神经网络的红外目标检测算法
Infrared Target Detection Algorithm Based on Target Feature Distribution Enhanced Convolutional Neural Network
丁胜男 1李威 1蔡立明 1李蒙 1胡常青1
作者信息
- 1. 航天时代(青岛)海洋装备科技发展有限公司,青岛266237;北京航天控制仪器研究所,北京100039
- 折叠
摘要
为了实现对水上红外弱小目标的探测,并减少由红外图像信噪比低、目标与背景的红外特征差异小等问题对检测结果的影响,通过结合红外图像的成分、纹理及目标形状特征,提出了基于目标特征分布增强卷积神经网络的红外目标检测(TFD_CNN)算法.该算法包含目标特征分布学习与深度神经网络,具备滤除红外图像中噪声的能力,并深度挖掘红外图像中目标的边缘、纹理及形状信息,提升了卷积神经网络的分类精度.通过与4种算法进行实验对比,TFD_CNN算法分类准确率为96％,高于其他算法.结果表明:TFD_CNN算法具备对红外图像中落水人员与船只的分类能力.
Abstract
In order to achieve the detection of infrared weak small maritime targets and reduce the impact of low sig-nal-to-noise ratio of infrared images and small differences in infrared features between targets and backgrounds on the detec-tion results, a infrared target detection algorithm based on target feature distribution enhanced convolutional neural network ( TFD_CNN) is proposed by combining the composition, texture, and target shape features of infrared images. The algo-rithm includes target feature distribution learning and deep neural networks, which has the ability to filter out noise in infra-red images and deeply mines the edge, texture and shape information of the target in the infrared image to improve the clas-sification accuracy of the convolutional neural network. Through the experimental comparison with four other algorithms, the classification accuracy of the TFD_CNN algorithm is 96％, which is higher than other algorithms. The results show that the TFD_CNN algorithm has the ability to classify drowning persons and ships in the infrared image.
关键词
红外弱小目标/目标特征分布学习/深度学习/目标分类Key words
infrared weak small target/target feature distribution learning/deep learning/object classification引用本文复制引用
基金项目
国家自然科学基金(61971153)
国家自然科学基金(62271159)
出版年
2024
NETL
NSTL
万方数据