Abstract
In recent years, Convolutional Neural Network (CNN) has been widely applied in speech/image/video recognition and classification. Although the results achieved are so impressive, CNN architecture is becoming more and more complex since CNN includes more layers to achieve better performance. In this paper, we developed a new CNN structure with several parallel CNNs and a Back-Propagation Neural Network (BPNN). The parallel CNNs can have the same or different numbers of layers. The outputs of the CNNs are the inputs of a fully connected BPNN. The structure of the proposed model can reduce the complexity of CNN by reducing the total number of CNN layers while the performance of feature extraction can be improved. The proposed model was validated based on CIFAR-10, CIFAR-100, and MNIST datasets and the achieved performance of the model is promising.