卷积神经网络在目标检测中的应用及FPGA实现

Application of convolutional neural network in target detection and FPGA implementation

针对传统目标检测算法在复杂背景条件下的对红外弱小移动目标的检测能力弱,虚警率高等问题,提出了一种基于卷积神经网络的目标检测方法,分析了卷积神经网络的结构、特点,将卷积神经网络应用到红外弱小目标检测领域,选择卷积神经网络模型,学习训练学习出合适的模型参数,并将算法在以FPGA为核心的硬件平台上进行移植。实验表明,本文的算法实时性好,硬件移植工作量小,在复杂背景下能够得到目标掩码信息、有效检出目标。

Aiming at the problem that the traditional target detection algorithm has weak detection ability and high false alarm rate under complex background conditions,a target detection method based on convolutional neural network is proposed.The structure and characteristics of convolutional neural network are analyzed.The convolutional neural network is applied to the infrared dim small targets detection,the convolutional neural network model is selected,the appropriate model parameters are learned through training,and the algorithm is transplanted on the hardware platform with FPGA as the core.Experiments show that the proposed algorithm has good real-time performance and small hardware migration workload,and can obtain the target mask information and effectively detect the target under complex background.