Multiple classifier for concatenate-designed neural network

This article introduces a multiple classifier method to improve the performance of concatenate-designed neural networks, such as ResNet and DenseNet, with purpose alleviate pressure on final classifier. We give design classifiers, which collects features produced between network sets, present constituent layers activation function for calculate classification score each use L2 normalization obtain instead Softmax normalization. also determine conditions that can enhance convergence. As result, proposed classifiers are able accuracy in experimental cases significantly, show not only has better than original models, but produces faster Moreover, our general be applied all related models.