if the input size is an image of shape (32, 32, 1), and we apply 32 convolution 2d layers of size (3, 3), how many parameters would be needed?