An image contains three channels, namely the RGB channels. In the three-channel convolution by accumulating three convolution result is obtained.
CNN layers fully connected in a convolution kernel size is the size of the feature map. Such feature is 3 * 3, then the connection layer full convolution kernel size of 3 * 3.
FCN is the last of three on CNN fully connected layers into a full convolution layer. In fact, the difference between the two is the convolution kernel of a different size. feature map is no longer the output size of 1 * 1.
The following is an example of my own writing, for your reference, if any mistakes welcome that.