摘要
随着深度学习网络的不断发展,卷积神经网络在图像识别与处理领域的正确率已达到甚至超越人类水平。但是,越来越复杂的网络结构导致庞大的计算模型体积和计算量,不利于模型的移植利用。基于此,分别介绍了网络压缩加速的典型方法并进行比较,在保证算法准确率损失最少的前提下,尽可能使算法具有可移植性,充分体现卷积神经网络算法的应用价值。
With the continuous development of deep learning network, the accuracy of convolutional neural network in image recognition and processing has reached or even surpassed human level. However, more and more complex network structure leads to huge computing model volume and computation, which is not conducive to the transplantation and utilization of the model. Based on this, the typical methods of network compression and acceleration are introduced and compared. On the premise of guaranteeing the least loss of accuracy, the algorithm can be transplanted as far as possible, which fully reflects the application value of convolutional neural network algorithm.
引文
[1]He Y,Zhang X,Sun J.Channel Pruning for Accelerating Very Deep Neural Networks[C].//International Conference on Computer Vision,2017.
[2]陈伟杰.卷积神经网络的加速及压缩[D].广州:华南理工大学,2017:81.
[3]Qin Z,Zhang Z,Chen X,et al.FD-MobileNet:Improved MobileNet with a Fast Downsampling Strategy[C].//2018 25th IEEE International Conference on Image Processing,2018.
[4]毕鹏程,罗健欣,陈卫卫.轻量化卷积神经网络技术研究[J].计算机工程与应用,2019(10):14.
[5]Hinton G,Vinyals O,Dean J.Distilling the Knowledge in a Neural Network[J].Computer Science,2015,14(7):38-39.