Why do we consider number of channels of previous layer in calculating trainable parameters ?(LeNet-5)

Question

According to my understanding - Trainable parameters in CNN network is nothing but the parameters which get changed during back propagation.