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...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。展开更多
针对复杂背景下红外场景对比度低、特征不足、细节不清而导致的目标检测效率低的问题,在YOLOv5s模型基础上通过创建TCC(two-way convolution and Concat)模块并引入华为Ghost模块,提出了一种基于改进YOLOv5s模型的红外弱小目标检测方法...针对复杂背景下红外场景对比度低、特征不足、细节不清而导致的目标检测效率低的问题,在YOLOv5s模型基础上通过创建TCC(two-way convolution and Concat)模块并引入华为Ghost模块,提出了一种基于改进YOLOv5s模型的红外弱小目标检测方法。首先,结合红外图像的低级语义特征,采取二路卷积和多尺度思想创建了TCC模块,提升了特征提取的全面性;接着,为进一步简化网络结构、减少网络参数量,引入轻量化Ghost模块改进了SPP池化层和CSP2卷积网络;最后,以无人机为实验对象,构建了白天和夜间不同背景条件下的红外弱小目标数据集,实验验证了本文改进算法的有效性。结果表明:改进后的YOLOv5s模型在较少损失帧频的情况下,检测精度提升了1.34%,平均精度均值(mean average precision, mAP)提升了2.26%,优于YOLOv4-tiny和YOLOv7-tiny两种轻量化模型,并与YOLOv8s模型精度相当,但模型参数量仅为YOLOv8s模型的53%,完全可以满足嵌入式设备部署的需求。展开更多
在军事作战领域中,准确的目标识别是确保诸如预警系统、拦截导弹、侦察装备及远程打击武器等各类军事资产能够充分发挥其战术与战略效能的核心要素。然而,在复杂背景的干扰下,通过雷达侦测、光电侦测、电磁频谱侦测等常规侦测手段已经...在军事作战领域中,准确的目标识别是确保诸如预警系统、拦截导弹、侦察装备及远程打击武器等各类军事资产能够充分发挥其战术与战略效能的核心要素。然而,在复杂背景的干扰下,通过雷达侦测、光电侦测、电磁频谱侦测等常规侦测手段已经难以满足现代战场环境下对于导弹、无人机等小目标的预警监测需求。本文针对当前战场环境中无人机等弱小空中目标监测与识别的紧迫需求,设计了一种地对空红外弱小目标自主发现设备。设备由电源模块、图像采集模块、嵌入式算法处理模块、显示模块、嵌入式算法处理模块组成,成本低廉、小巧轻便、便携性强。同时,设备引入最新的DNA-Net模型来进行红外小目标识别。鉴于传统侦测手段如雷达、光电及电磁侦测在便携性、隐蔽性上的局限,以及难以满足复杂环境中的实时监测挑战,本研究聚焦于利用红外成像技术与深度学习算法的结合,以提高弱小目标的发现能力。In the field of military operations, accurate target identification is a core element to ensure that military assets such as early warning systems, interceptor missiles, reconnaissance equipment and long-range strike weapons can achieve their full tactical and strategic effectiveness. However, under the interference of complex background, conventional detection methods such as radar detection, photoelectric detection and electromagnetic spectrum detection have been difficult to meet the needs of early warning and monitoring of small targets such as missiles and UAVs in modern battlefield environment. Aiming at the urgent need of monitoring and recognition of small and small targets such as unmanned aerial vehicles (UAVs) in the current battlefield environment, this paper designs a surface-to-air infrared small and small targets autonomous detection equipment. The device is composed of power module, image acquisition module, embedded algorithm processing module, display module and embedded algorithm processing module, with low cost, compact and portable. At the same time, the device introduces the latest DNA-Net model for infrared small target recognition. In view of the limitations of traditional detection methods such as radar, photoelectric and electromagnetic detection in portability and concealability, as well as the difficulty in meeting the challenges of real-time monitoring in complex environments, this research focuses on the combination of infrared imaging technology and deep learning algorithm to improve the detection ability of dim targets.展开更多
文摘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。
文摘针对复杂背景下红外场景对比度低、特征不足、细节不清而导致的目标检测效率低的问题,在YOLOv5s模型基础上通过创建TCC(two-way convolution and Concat)模块并引入华为Ghost模块,提出了一种基于改进YOLOv5s模型的红外弱小目标检测方法。首先,结合红外图像的低级语义特征,采取二路卷积和多尺度思想创建了TCC模块,提升了特征提取的全面性;接着,为进一步简化网络结构、减少网络参数量,引入轻量化Ghost模块改进了SPP池化层和CSP2卷积网络;最后,以无人机为实验对象,构建了白天和夜间不同背景条件下的红外弱小目标数据集,实验验证了本文改进算法的有效性。结果表明:改进后的YOLOv5s模型在较少损失帧频的情况下,检测精度提升了1.34%,平均精度均值(mean average precision, mAP)提升了2.26%,优于YOLOv4-tiny和YOLOv7-tiny两种轻量化模型,并与YOLOv8s模型精度相当,但模型参数量仅为YOLOv8s模型的53%,完全可以满足嵌入式设备部署的需求。
文摘在军事作战领域中,准确的目标识别是确保诸如预警系统、拦截导弹、侦察装备及远程打击武器等各类军事资产能够充分发挥其战术与战略效能的核心要素。然而,在复杂背景的干扰下,通过雷达侦测、光电侦测、电磁频谱侦测等常规侦测手段已经难以满足现代战场环境下对于导弹、无人机等小目标的预警监测需求。本文针对当前战场环境中无人机等弱小空中目标监测与识别的紧迫需求,设计了一种地对空红外弱小目标自主发现设备。设备由电源模块、图像采集模块、嵌入式算法处理模块、显示模块、嵌入式算法处理模块组成,成本低廉、小巧轻便、便携性强。同时,设备引入最新的DNA-Net模型来进行红外小目标识别。鉴于传统侦测手段如雷达、光电及电磁侦测在便携性、隐蔽性上的局限,以及难以满足复杂环境中的实时监测挑战,本研究聚焦于利用红外成像技术与深度学习算法的结合,以提高弱小目标的发现能力。In the field of military operations, accurate target identification is a core element to ensure that military assets such as early warning systems, interceptor missiles, reconnaissance equipment and long-range strike weapons can achieve their full tactical and strategic effectiveness. However, under the interference of complex background, conventional detection methods such as radar detection, photoelectric detection and electromagnetic spectrum detection have been difficult to meet the needs of early warning and monitoring of small targets such as missiles and UAVs in modern battlefield environment. Aiming at the urgent need of monitoring and recognition of small and small targets such as unmanned aerial vehicles (UAVs) in the current battlefield environment, this paper designs a surface-to-air infrared small and small targets autonomous detection equipment. The device is composed of power module, image acquisition module, embedded algorithm processing module, display module and embedded algorithm processing module, with low cost, compact and portable. At the same time, the device introduces the latest DNA-Net model for infrared small target recognition. In view of the limitations of traditional detection methods such as radar, photoelectric and electromagnetic detection in portability and concealability, as well as the difficulty in meeting the challenges of real-time monitoring in complex environments, this research focuses on the combination of infrared imaging technology and deep learning algorithm to improve the detection ability of dim targets.
