多视觉传感器协同弱小目标检测  

Multi Vision Sensor Coordination Dimand Small Target Cooperative Detection

作　　者：王田 程嘉翔[1] 刘克新[2,3,4] 王薇 吕金虎 WANG Tian;CHENG Jiaxiang;LIU Kexin;WANG Wei;LYU Jinhu(Institute of Artificial Intelligence,Beihang University,Beijing 100191,China;School of Automation Science and Electrical Engineering,Beihang University,Beijing 100191,China;State Key Laboratory of Complex&Critical Software Environment,Beihang University,Beijing 100191,China;Zhongguancun Laboratory,Beijing 100191,China)

机构地区：[1]北京航空航天大学人工智能研究院,北京100191 [2]北京航空航天大学自动化科学与电气工程学院,北京100191 [3]北京航空航天大学复杂关键软件环境全国重点实验室,北京100191 [4]中关村实验室,北京100191

出　　处：《指挥与控制学报》2024年第1期9-18,共10页Journal of Command and Control

基　　金：国家自然科学基金(62032016,61972016);北京市科技新星基金(20220484106,20230484451)资助。

摘　　要：多视觉传感器协同对空实现全区域覆盖的弱小目标检测,在近距离防空领域中具有重要意义。现有的全区域覆盖方法存在覆盖率低、随机性差等问题,弱小目标检测算法存在模型大、定位及分类准确性低等问题。提出了一种高效的对空全区域覆盖算法和轻量级弱小目标检测算法,通过结合最大面积优先法和最小曼哈顿离法改善存在覆盖死角和随机性差等问题。提出密集通道扩展网络(dense and channel expand network,DCENet)模型,基于轻量级稠密拼接和自适应尺寸通道扩展方法,在弱小目标数据集上获得了比原算法更有竞争力的平均精度结果。Multi vision sensor coordination can realize the dim target detection by covering all areas in the air,which is of great significance in the field of close air defense.The existing full area coverage methods have problems such as low coverage rate and poor randomness,while the dim target detection algorithms have problems such as big model,low accuracy of location and classification.An efficient full area coverage algorithm and a lightweight dim target detection algorithm covering all area in the air are proposed.By combining the maximum area priority method and the minimum Manhattan distance method,the problems of dead corner coverage and poor randomness are improved.A dense channel expand network(dense and channel Expand network,DCENet)model is proposed.Based on the lightweight dense stitching and adaptive size channel expansion method,the average accuracy results on dim target data sets are more competitive than that of the original algorithm.

关 键 词：协同目标检测 全区域覆盖 弱小目标检测 轻量级稠密拼接 

分 类 号：E926.4[军事—军事装备学] TP391.41[兵器科学与技术—武器系统与运用工程]