摘要
2006年深度学习的提出为机器学习领域带来新的革命,深度学习的成功不仅依赖于理论和模型上的突破,也离不开大数据环境下的海量训练样本以及不断革新的先进计算技术。在GPU被应用于科学计算之前,神经网络特别是大型神经网络的训练时间往往令人生畏。GPU特别适应于并行计算的特性给神经网络的训练速度带来数十倍的提升。开源的GPU计算框架也不断地推陈出新,推动深度学习在各方面的应用,Caffe就是其中的一种。由于简单易用、性能强大,Caffe框架受到了广泛的认可。利用Caffe框架对印章类型进行识别,所采用的两种模型都取得极好的实验效果,对印章的自动识别提供新的参考。
The proposal of deep learning in 2006 brought a revolution to the field of machine learning. Huge amount of available training data and new computational techniques lie in the key position of deep learning＇s success. GPU has the very feature of parallel computation, and this feature can speed up the training of neural networks by a factor of more than 40. Many open source frameworks of GPU computation have been proposed during the years. Among them, Caffe is one of good reputation for its many merits. Utilizes the Caffe Framework for the task of Seal type recognition, both models used have shown very good performance.
基金
国家自然科学基金资助项目(61303015)
四川省科技计划项目(No.2014GZ0005-5)
