人工智能培训

搜索

人工智能论文:同类CNN / LSTM的有限大小集合,用于高性能单词分类(A limited-size ensemble of homogeneous CNN/

[复制链接]
nite_p007 发表于 2019-12-9 14:47:49 | 显示全部楼层 |阅读模式
nite_p007 2019-12-9 14:47:49 383 0 显示全部楼层
人工智能论文:同类CNN / LSTM的有限大小集合,用于高性能单词分类(A limited-size ensemble of homogeneous CNN/LSTMs for high-performance  word classification)近年来,长短期记忆神经网络(LSTM)已非常成功地应用于手写文本识别中的问题,但是,它们的优势更多地位于处理可变长度序列中,而不是在处理图像图案的几何可变性方面。 LSTM的最佳结果通常基于对网络实例集合的大规模培训。在本文中,使用端到端卷积LSTM神经网络来处理几何变异和序列变异。我们表明,通过使用适当的编码方案和适当的投票方案仅对五个这样的网络使用适当的数据增强,就可以在公共基准集上达到高性能。这些网络具有相似的架构(卷积神经网络(CNN):五层,双向LSTM(BiLSTM):三层,其后是连接主义的时间分类(CTC)处理步骤)。该方法假定不同比例的输入图像和不同的特征图大小。我们使用两个数据集来评估算法的性能:标准基准RIMES数据集(法语)和历史手写数据集KdK(荷兰语)。通过RIMES的单词识别测试获得的最终性能为96.6%,明显优于其他最新方法。在KdK数据集上,我们的方法也显示出良好的结果。提议的方法已部署在Monksearch引擎中以用于历史笔迹收藏。
In recent years, long short-term memory neural networks (LSTMs) have beenapplied quite successfully to problems in handwritten text recognition.However, their strength is more located in handling sequences of variablelength than in handling geometric variability of the image patterns.Furthermore, thebest results for LSTMs are often based on large-scale trainingof an ensemble of network instances.In this paper, an end-to-end convolutionalLSTM Neural Network is used to handle both geometric variation and sequencevariability.We show that high performances can be reached on a commonbenchmark set by using proper data augmentation for just five such networksusing a proper coding scheme and a proper voting scheme.The networks havesimilar architectures (Convolutional Neural Network (CNN): five layers,bidirectional LSTM (BiLSTM): three layers followed by a connectionist temporalclassification (CTC) processing step).The approach assumes differently-scaledinput images and different feature map sizes.Two datasets are used forevaluation of the performance of our algorithm: A standard benchmark RIMESdataset (French), and a historical handwritten dataset KdK (Dutch).Finalperformance obtained for the word-recognition test of RIMES was 96.6%, a clearimprovement over other state-of-the-art approaches.On the KdK dataset, ourapproach also shows good results.The proposed approach is deployed in the Monksearch engine for historical-handwriting collections.人工智能论文:同类CNN / LSTM的有限大小集合,用于高性能单词分类(A limited-size ensemble of homogeneous CNN/LSTMs for high-performance  word classification)
URL地址:https://arxiv.org/abs/1912.03223     ----pdf下载地址:https://arxiv.org/pdf/1912.03223    ----人工智能论文:同类CNN / LSTM的有限大小集合,用于高性能单词分类(A limited-size ensemble of homogeneous CNN/LSTMs for high-performance  word classification)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

nite_p007当前离线
新手上路

查看:383 | 回复:0

快速回复 返回顶部 返回列表