IMPROVING FEATURE MAPS IN EARLY LAYERS OF CONVOLUTIONAL NEURAL NETWORKS USING OTSU METHOD | ||||
International Journal of Intelligent Computing and Information Sciences | ||||
Article 3, Volume 16, Issue 2, April 2016, Page 37-45 PDF (882.9 K) | ||||
Document Type: Original Article | ||||
DOI: 10.21608/ijicis.2018.10905 | ||||
View on SCiNiTO | ||||
Authors | ||||
A Al-furas1; M AL-dosuky1; Taher Hamza2 | ||||
1Faculty of Computer and Information,Mansoura University, Egypt. | ||||
2Computer Science Department Faculty of Computer and Information Sciences, Mansoura University - Egypt | ||||
Abstract | ||||
Abstract: A novel deep architecture Thresholding Convolution Neural Network (ThCNN) progresses in this paper; Which is a simple and effective method to regularizing features map in the early layers of Convolution Neural Network(CNN). One of the issues identified with deep learning is the features in early layers that robustness and discriminativeness. In this paper, we compute the optimal global threshold to determine the features that are passed to the next layers. We then evaluate ThCNN on an MNIST dataset comparing it CNN by applying multiple trained models. It yield decent accuracy compared to traditional CNN. It gives a 99.5% | ||||
Statistics Article View: 170 PDF Download: 394 |
||||