摘要
近年来,卷积神经网络在目标检测、图像语义分割和图像识别领域取得了一系列重大突破性的成果。但是随着检测率的提升,网络结构也在向着更复杂的方向发展。为解决卷积神经网络结构复杂,样本的检测时间过长的问题,论文提出了一种通过特征图之间的差异性对卷积核数目进行优化的方法,通过计算得到最优卷积核数目,降低网络复杂度,从而加快样本检测的速度,提高泛化能力。实验结果表明,该方法在保证准确率的前提下,提升了检测速度。
In recent years,convolutional neural network has made a series of major breakthrough in target detection,image semantic segmentation and image recognition.But with the improvement of the detection rate,the network structure is becoming more and more complex.In order to solve the problem of complex structure and much time in detecting.In this paper,a method is proposed to optimize the number of convolution kernels by the difference between the feature maps.By calculating the difference between the feature maps,in order to reduce network complexity,increase the detection speed,and improve generalization ability.Experimental results show that the method will increase the detection speed and ensure the accuracy rate.
作者
刘晨
曲长文
周强
李智
LIU Chen;QU Changwen;ZHOU Qiang;LI Zhi(Department of Electronic and Information Engineering,Naval Aeronautical Engineering Institute,Yantai 264001)
出处
《舰船电子工程》
2017年第5期36-40,共5页
Ship Electronic Engineering
关键词
深度学习
卷积神经网络
卷积核
特征图
手写数字识别
deep learning
convolutional neural network
convolutional kernel
characteristic pattern
handwritten numeral recognition
