摘要
为解决无人机带来的安全隐患与隐私侵犯等问题,提出选择性坐标注意力下红外图像无人机目标检测方法。基于选择性坐标注意力机制,通过非对称卷积核在不同方向上捕捉不同尺度和形状的特征,将无人机特征的行列位置信息进行编码,动态地调整不同位置特征的权重,强化关键区域的特征表示。将多个红外图像输入YOLOv5网络中进行训练和处理后,在主干网络中经卷积操作后嵌入选择性坐标注意力机制,实现红外图像无人机目标特征精确提取。采用GIoU作为损失函数,优化预测框的位置和大小,实现红外图像无人机目标精准检测。实验结果表明:该方法对大小不同的无人机目标均能实现准确且快速的定位与检测,能够保持较高的检测精度。
This paper proposes an unmanned aerial vehicle(UAV)object detection method for infrared images based on selective coordinate attention in order to avoid the security risks and privacy violations caused by UAVs.On the basis of the selective coordinate attention mechanism,the asymmetric convolutional kernels are used to capture features of different scales and shapes in different directions.The row and column position information of UAV features is encoded,the weights of different position features are adjusted dynamically,and the feature representations of key regions are strengthened.After inputting multiple infrared images into the YOLOv5 network for training and processing,a selective coordinate attention mechanism is embedded in the backbone network after convolution operation,so as to achieve accurate feature extraction of UAV objects in infrared images.The GIoU is taken as the loss function to optimize the position and size of the prediction box,so as to achieve accurate detection of UAV objects in infrared images.The experimental results show that the method can locate and detect UAV objects of different sizes accurately and quickly,and can maintain high detection accuracy.
作者
吴茜
魏晶鑫
陈中举
WU Xi;WEI Jingxin;CHEN Zhongju(School of Computer Science,Yangtze University,Jingzhou 434023,China)
出处
《现代电子技术》
北大核心
2025年第7期43-47,共5页
Modern Electronics Technique
基金
中国高校产学研创新基金:新一代信息技术创新项目(2023IT269)。
