基于深度学习及FPGA的装备目标检测研究

Missile target detection based on Deep learning and its implementation in an FPGA

本文应用深度学习技术实现海天背景下基于可见光、红外方式成像的舰船及角反、烟幕干扰的目标检测,这也是反舰导弹作战使用的关键技术之一。采集的可见光与红外成像目标检测数据集涵盖实施典型干扰下的态势场景,贴近实战;结合四种不同的目标检测机制,选取YOLOV3,Faster R-CNN、SSD及CenterNet四种典型模型分别进行训练与验证,通过对比分析进一步提高弱小目标、复杂干扰态势的的检测,可以实现端到端的高精度装备目标检测模型。在确保精度的前提下基于现场可编程门阵列(FPGA)进行软硬件协同设计,通过对比分析选定基于VitisAI的实施方案,经过模型的量化、编译与优化,可在保证检测效率的前提下快速实现模型的小型化部署,便于进行装备移植。研究结果表明,该研究内容可有效提高现役反舰导弹目标检测的准确率。

This paper proposes the use of Deep learning technology for target detection, angle counter, and smoke jamming through visible-light and infrared imaging with the sea or the sky as background, where this is a key technology for anti-ship missile combat. A dataset for target detection consisting of visible-light and infrared images was collected to cover scenarios involving typical interference to simulate actual combat situations. Four target detection mechanisms were used to train and verify four typical models: YOLOv3, Faster R-CNN, SSD,and CenterNet. Comparative analysis was used to improve the detection of weak and small targets in case of complex jamming to develop a model for end-to-end high-precision target detection equipment. On the premise of ensuring high accuracy, a Field Programmable Gate Array(FPGA) was designed using the aforementioned software and an implementation scheme based on Vitis AI was used. After the quantification, compilation, and optimization of the model, a miniaturized version of it was deployed to show that our target detection framework can be used in miniature systems and embedded into different types of equipment without compromising accuracy. The results show that the proposed method can improve the accuracy of target detection by anti-ship missiles.