基于改进SSD的航拍飞机目标检测方法  

Aircraft target detection method based on improved SSD

作　　者：李静[1,2] 喻佳成[1] 张灵灵 LI Jing;YU Jia-cheng;ZHANG Ling-ling(College of Electronic Information Engineering,Xi'an Technological University,Xi'an 710021,China;College of Ordnance Science and Technology,Xi'an Technological University,Xi'an 710021,China)

机构地区：[1]西安工业大学电子信息工程学院,陕西西安710021 [2]西安工业大学兵器科学与技术学院,陕西西安710021

出　　处：《液晶与显示》2023年第1期128-137,共10页Chinese Journal of Liquid Crystals and Displays

基　　金：西安工业大学动态测试与智能信息处理创新团队基金(No.401-302021605);国防基础科研项目(No.JCKY2016606b001)。

摘　　要：针对航拍图像中对于小尺度的飞机目标出现漏检、错检的问题,在SSD(Single Shot MultiBox Detector)模型的基础上提出了一种改进SSD的航拍图像目标检测模型。首先,针对SSD模型中浅层特征图中缺乏语义、细节信息的问题,设计了一种特征融合机制,通过添加细节信息补充特征层和添加由递归反向路径获得的语义信息补充特征层来丰富浅层特征图的语义、细节信息。然后,针对SSD模型对通道以及空间信息的关注能力不足的问题,引入了结合通道和空间的混合注意力模块来提高模型整体的关注能力。最后,针对SSD模型中先验框与小尺度目标不匹配的问题,对先验框的比例进行了调整。使用自制的航拍图像数据集进行验证,结果表明改进后的模型检测精度为95.7%,相较于原模型提高了7.5%,检测速度达到30.8 FPS。For the problem of omission and misdetection of small-scale aircraft targets in aerial images,an improving SSD object detection model is proposed based on SSD(Single Shot MultiBox Detector)model. Firstly,in view of the lack of semantic and detailed information in the shallow feature map in the SSD model,a feature fusion mechanism is designed to enrich the semantic and detailed information of the shallow feature layer by adding the supplementary feature layer obtained from the recursive reverse path.Then,to address the problem of the SSD model to focus on the channels and spatial information,a hybrid attention module combining channels and space is introduced to improve the overall attention ability of the model. Finally,the proportion of prior boxes is adjusted for the problem of mismatch to small-scale targets in the SSD model. The self-made aerial images data set is used for verification. The results show that the improved algorithm accuracy is 95. 7%,which is 7. 5% higher than original SSD algorithm,and the detection speed is 30. 8 FPS.

关 键 词：目标检测 特征融合 注意力机制 航拍图像 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]