Stacking ensemble model of deep learning and its application to Persian/Arabic handwritten digits recognition

Highlights：

• A new stacking ensemble deep learning model is proposed.

• Proposed model is based on conventional neural network in first layer and bidirectional long short term memory in second layer as meta classifier.

• Bidirectional long short term memory learn output pattern of convolutional neural network layer and improve misclassification data in first layer.

• The model is benchmarked on dataset which is contain 80000 real images data. The role of the meta classifier is confirmed by the result of confusion matrix.

• The results on datasets confirm the performance of proposed model in practice and it has achieved better results than other methods.

摘要

•A new stacking ensemble deep learning model is proposed.•Proposed model is based on conventional neural network in first layer and bidirectional long short term memory in second layer as meta classifier.•Bidirectional long short term memory learn output pattern of convolutional neural network layer and improve misclassification data in first layer.•The model is benchmarked on dataset which is contain 80000 real images data. The role of the meta classifier is confirmed by the result of confusion matrix.•The results on datasets confirm the performance of proposed model in practice and it has achieved better results than other methods.

论文评审过程：Received 5 January 2021, Revised 1 March 2021, Accepted 8 March 2021, Available online 10 March 2021, Version of Record 18 March 2021.