Cross Convolutional Neural Networks

摘要

Using convolutional neural networks (CNNs) to classify images is one of the research hotspots in recent years. In order to achieve a good classification effect, the learning parameters of the convolutional neural network should be reduced as much as possible. Therefore, a new operation, cross convolutional, is introduced in this paper, which uses the principle of target calibration in YOLO algorithm and the transformation function. It generalizes convolution, reduces the volume of the model, but without affecting the ability to capture high-dimensional features. The experiments show that as the network depth increases, the cross convolutional neural network can not only reduce more learning parameters, but also achieve better classification results than baseline CNN.