一种基于无人机探测图像区块信息的弱小目标检测算法  

Weak Object Detection Algorithm Based on High Resolution Remote Sensing Image of UAV Platform

作　　者：李楚晨 唐善军 赵冰青 LI Chuchen;TANG Shanjun;ZHAO Bingqing(Shanghai Electro-Mechanical Engineering Institute,Shanghai 201109,China)

机构地区：[1]上海机电工程研究所,上海201109

出　　处：《空天防御》2025年第1期41-47,共7页Air & Space Defense

摘　　要：无人机平台高分辨率遥感图像中的目标信息是军事行动规划的重要参考依据。然而高分辨率遥感图像存在尺寸较大、背景信息混杂的问题,导致目标信息检测结果常存在大量虚警以及错误识别;不仅如此,由于图像与目标尺寸之间存在巨大差异,因此在大量背景像素中探测微小目标,对于检测算法的精确度有着更高的要求。为了解决上述问题,实现准确目标检测,本文提出了一种基于无人机遥感图像区块信息的弱小目标检测算法。算法首先使用区块信息以及微小神经网络结构来缓解大尺寸图像带来的处理困难、速度较慢的问题;其次使用全局注意力机制对于可能出现的虚警进行抑制;最终将检测器与分类器进行信息互通,从而同时提高两者的能力,达到提升模型性能的效果。所提检测算法在大尺寸遥感图像数据集上进行实验,结果表明,检测结果中假阳性个数明显减少,并且准确度大幅提升。该结果证明本文所提方法的有效性。The target information from the high-resolution remote sensing images using unmanned aerial vehicle(UAV)platforms is critical to military action planning.However,the large size and mixed background information of high-resolution remote sensing images can cause a lot of false alarms and misidentifications in the target information detection results.In addition,due to the huge difference between the image and the target size,detecting small targets in a large number of background pixels must employ a higher accuracy detection algorithm.To solve the above problems,this paper proposed a tiny target detection algorithm utilizing the block information of UAV remote sensing images.The model first alleviated the processing difficulties and slow speed caused by large image sizes using block information and a small neural network structure.Then,the global attention mechanism was applied to suppress possible false alarms.Finally,the detector and the classifier information during the training process was exchanged to improve the capabilities of both at the same time.The proposed detection algorithm was verified on a large-size remote sensing image dataset.The results show that the number of false positives in the detection results is significantly reduced and the accuracy is greatly improved,demonstrating the effectiveness of the proposed method.

关 键 词：无人机探测 高分辨率遥感图像 目标检测 弱小目标 注意力机制 

分 类 号：V279[航空宇航科学与技术—飞行器设计]