摘要
针对红外无人机目标识别过程中特征信息较少、特征丢失严重、识别准确率低等问题,提出一种基于YOLOv7的红外无人机目标检测方法。通过引入注意力机制来增强目标区域的特征表达能力,提升图像的空间信息含量。采用改进的串行连接方式将通道注意力模块与空间注意力模块连接,在结合了目标通道特征信息和空间特征信息的同时,改进结构降低了通道注意力对红外图像识别的负面影响,可以更好地实现对红外目标的特征加强作用。选择基于角度向量回归的SIoU损失函数作为边框损失函数,进一步提升了模型的收敛性和检测精度。实验结果表明,改进的算法模型推理速度达到了43帧/s,准确率为95.4%,召回率为87.3%,mAP为96.1%,在红外无人机检测任务中取得了更好的检测效果。
To solve the problems of inadequate feature information,serious feature loss and low recognition accuracy in the process of infrared UAV target recognition,an infrared UAV target detection method based on YOLOv7 is proposed.By introducing the attention mechanism,the feature representation ability of the target region is enhanced and the spatial information content of the image is improved.The improved serial connection mode is used to connect the channel attention module to the spatial attention module.While combining the channel feature information with the spatial feature information,the improved structure reduces the negative impact of the channel attention on infrared image recognition,and can better realize the feature strengthening of the infrared target.The SIoU loss function based on angle vector regression is selected as the frame loss function,which further improves the convergence and detection accuracy of the model.The experimental results show that the reasoning speed of the improved algorithm model reaches 43 frames per second,the accuracy is 95.4%,the recall rate is 87.3%,and the mAP is 96.1%.Better results are obtained in the infrared UAV detection task.
作者
梁晓
李俊
LIANG Xiao;LI Jun(Guilin University of Electronic Technology,College of Electronic Engineering and Automation,Guilin 541000,China;Guilin University of Electronic Technology,College of Computer and Information Security,Guilin 541000,China)
出处
《电光与控制》
CSCD
北大核心
2023年第12期38-43,92,共7页
Electronics Optics & Control
基金
国家自然科学基金(61866009)。
