Feature Mining: A Novel Training Strategy for Convolutional Neural Network

In this paper, we propose a novel training strategy named Feature Mining for convolutional neural networks (CNNs) that aims to strengthen the network’s learning of local features. Through experiments, found different parts feature contain semantics, while network will inevitably lose large amount information during feedforward propagation. order enhance features, divides complete into two complementary and reuses divided make capture information; call steps Segmentation Reusing. is parameter-free method with plug-and-play nature can be applied any CNN model. Extensive experiments demonstrated wide applicability, versatility, compatibility our method.