Convolutional Neural Network for Image Feature Extraction Based on Concurrent Nested Inception Modules
Zhengyan Wang, Junfeng Chen, Xiaolin Wang · 2019
Nowadays, convolutional neural networks are getting deeper and more complex to extract and learn features at multiple levels of abstraction. However, large networks take time to be trained, and the training process is not that easy. In this paper, a new architecture based on Inception module is proposed. In each branch of the normal Inception module, we have nested numbers of secondary Inception module to further extract features on the input image and use the width of the net to expand the quantity of the features being extracted. This structure helps the net build a more comprehensive cognition on the image and get great performance. Multiple experiments and comparisons on MNIST, CIFAR-10 and SVHN datasets have proved the feasibility and performance of the structure, and the structure converges well when combined with deeper networks.