In the diagram, each neuron in the convolutional layer is connected to a 3×3 matrix of input neurons, color-coded for clarity, and those matrices overlap by one.