摘要
红外序列图像小目标在环境干扰下检测准确性不好,为了提高红外序列图像小目标检测能力,提出基于深度学习的红外序列图像小目标检测方法。构建红外序列小目标图像的三维成像模型,采用边缘特征分割和角点分布式提取方法进行红外序列小目标图像的多维尺度分解,构建目标图像的三维成像模型,采用灰度信息重构方法进行红外目标图像的像素序列重组,建立目标图像的三维轮廓特征分布集,结合模板特征匹配和超像素特征序列重构方法进行红外序列图像小目标序列重构,采用模糊信息度特征提取方法实现红外序列小目标图像的特征提取和优化检测,构建红外序列小目标图像的像素分布灰度像素集,通过深度学习算法进行红外序列图像小目标检测过程中的自适应寻优,实现红外序列图像小目标检测优化。仿真结果表明,采用该方法进行红外序列图像小目标检测的自适应性较好,特征分辨能力较强,提高了目标的检测精度。
The detection accuracy of infrared sequence image small target under environmental interference is not good.In order to improve the detection ability of infrared sequence image small target,an infrared sequence image small target detection method based on depth learning is proposed.The 3 D imaging model of infrared sequence small target image is constructed,the multi-dimensional scale decomposition of infrared sequence small target image is carried out by using edge feature segmentation and corner distributed extraction method,the 3 D imaging model of target image is constructed,the pixel sequence reconstruction method of infrared target image is used to reconstruct the pixel sequence of infrared target image,and the 3 D profile feature distribution set of target image is established.Combining template feature matching and super pixel feature sequence reconstruction,infrared sequence image small target sequence reconstruction is carried out.Fuzzy information degree feature extraction method is used to realize the feature extraction and optimization detection of infrared sequence small target image.The pixel distributed gray pixel set of infrared sequence small target image is constructed,and the adaptive optimization in the process of infrared sequence image small target detection is carried out by depth learning algorithm.The optimization of infrared sequence image small target detection is realized.The simulation results show that the method has good adaptability and feature resolution,and improves the detection accuracy of infrared sequence images.
作者
周贵华
许丽娟
周伟昌
ZHOU Guihua;XU Lijuan;ZHOU Weichang(Guangzhou University Songtian College,Guangdong 511310,China;Huashang College,Guangdong University of Finance&Economics,Guangdong 511300,China;School of Physics and Electronics,Hunan Normal University 410000,China)
出处
《激光杂志》
北大核心
2020年第12期61-64,共4页
Laser Journal
基金
国家自然科学基金(No.51102091)。
关键词
深度学习
红外序列图像
小目标
检测
像素
depth learning
infrared sequence image
small target
detection
pixel