摘要
为了有效提高无人机航拍图像的识别准确率,本文提出了一种基于深度残差网络的深度卷积神经网络模型。该模型在深度残差网络的基础上使用了随机化ReLU激励函数,能够使网络拥有更加快速的收敛速度,同时针对深度残差网络层数较深,且部分参数在梯度下降的传播过程中参与度不高的特点,利用随机失活有效降低整个网络训练量,进而提高网络的训练速度。经过实验验证,本文提出的图像识别算法相比几种经典深度卷积神经网络模型拥有更高的识别准确度。
In order to effectively improve the recognition accuracy of UAV aerial image, this paper puts for-ward a kind of convolutional neural network model based on deep residual net. This model based on deep residual net uses the randomized ReLU excitation function, which can make the network have more rapid convergence speed. For participation of some parameters in the spread of gradient descent process which is not high in deep nets, the model uses random dropout to reduce the amount of calculation, improving training speed of network. The image recognition model proposed in this paper is verified having higher identification accuracy comparing with several classical deep convolution neural network model by experiment.
出处
《计算机科学与应用》
2018年第10期1613-1619,共7页
Computer Science and Application
