Research on image classification based on HP — Net convolutional neural networks
Qiang Wang, Xiaojie Li, Canghong Shi · 2017
Based on the Caffe deep learning framework and the first convolution layer inversion operation, this paper presents a deep-learning framework, denoted as HP-Net, to achieve image classification problem, The HP-Net network consists of three convolution layers and max-pooling layers followed by three fully connected layers. The first convolution layer adopts the inversion operation to add the transmission of the effective feature information and uses a smaller convolution kernel to extract more texture features. Finally, the softmax classifier is employed to identify image classifications. In this method, we use a very efficient graphics processing unit implementation of the convolution operation to further reduce training time. The HP-Net model demonstrated better performance two real-world datasets than two related state-of-the-art approaches, CaffeNet and AlexNet.