1x1 Convolution

With the deepening of the number of layers, the receptive field gradually increases, and the extracted features contain more abstract semantic information, which is used to judge the image task.

In 2014, in Network in Network, in order to improve the abstract expression ability of features, 1x1conv used 1x1 convolution + ReLU operation to increase the nonlinearity of the network, so as to improve the nonlinear fitting ability of the network without increasing or even decreasing the network parameters. Thereby improving the classification effect of the network.