Convolutional Neural Network (CNN) is one of the most efficient Deep Neural Networks. The addition of more layers and neurons to the CNN increases its computational complexity. Even though CNNs are capable of solving many real time image recognition tasks flawlessly, it is also crucial to design optimum neural network architecture by reducing the associated memory and computational costs for resource critical applications. The proposed method optimizes a pre-trained CNN model for Traffic Sign Recognition by identifying and eliminating the redundant channels in fully connected layer of the neural network. The basis of the algorithm is that in a large neural network, the contribution of some of the neurons is negligible and can be eliminated without much effect on the overall performance. After eliminating the channels, the resulting model is retrained to compensate for the performance loss. The process of elimination and retraining is repeated until no more redundant channels are identified. The performance of the models so developed are further compared with the original model and evaluated based on accuracy and inference time. By removing 69% of the neurons in the fully connected layer, a compression rate of 2.85 was achieved and inference time got reduced by 97ms .The model so developed had accuracy slightly higher than the original model due to the retraining performed after each iteration.

Convolutional Neural Network, Network Pruning, Traffic Sign Recognition

[1]. Denil, M., Shakibi, B., Dinh, L., Ranzato, M., de Freitas, N.: Predicting parameters in deep learning. CoRR abs/1306.0543(2013)
[2]. Lin, M., Chen, Q., Yan, S.: Network in network. CoRR abs/1312.4400(2013)
[3]. Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S.E.,Anguelov, D., Erhan, D., Vanhoucke, V.,Rabinovich,A.: Going deeper with convolutions. CoRRabs/1409.4842(2014)
[4]. Hanson, S.J., Pratt, L.: Advances in neural information processing systems 1. Morgan Kaufmann Publishers Inc., San Francisco, CA, USA (1989) 177–185
[5]. Hassibi, B., Stork, D.G.: Second order derivatives for network pruning: Optimal brain surgeon In:Advances in Neural Information Processing Systems 5, [NIPS Conference], San Francisco, CA, USA, Morgan Kaufmann Publishers Inc. (1993) 164–171
[6]. HHan, S., Mao, H., Dally, W.J.: Deep compression: Compressing deep neural networks with pruning trained quantization and Huffman coding. arXiv preprint arXiv:1510.00149 (2015)
[7]. HHengyuan Hu, Rui Peng, Yu-Wing Tai, and Chi-Keung Tang. Network trimming: A data-driven neuron pruning approach towards efficient deep architectures. arXiv preprint arXiv:1607.03250, 2016.
[8]. YYihui He ; Xiangyu Zhang ; Jian Sun : Channel Pruning for Accelerating Very Deep Neural Networks (2017), IEEE International Conference on Computer Vision (ICCV)
[9]. Kaoutar Sefrioui Boujemaa, Ismail Berrada, Afaf Bouhoute: Traffic sign recognition using convolutional neural networks, 2017 International Conference on Wireless Networks and Mobile Communications (WINCOM)
[10]. GTSRB Dataset-http://dx.doi.org/10.1016/j.neunet.2012.02.016