Convolution calculation parameters

FIG assumed characteristics (feature maps) a convolution of the input layer number (input channels) is "n", the number of output characteristic diagram is a "m", the convolution kernel (kernel size) as "k". Suppose we are dealing with a parameter 2D convolution operation amount, the input layer corresponds to the convolution k * k * n, at the same time, since the number of output channels m a characteristic diagram, in order to map the output of the convolutional layer needs learning (k * k * n) * m parameters, but not ignore this error term (= wx + b is known from Z), due to the general operation of the convolution is a matrix operation, b thus having characteristics broadcasting, then the current convolution parameters need to calculate the final layer of two (k * k * n + b) * m.