Mastering Computer Vision with TensorFlow 2.x

Fully connected layers

Fully connected layers, also known as dense layers, connect every connected neuron in the current layer to every connected neuron in the previous layer by applying weight and bias to them. The vector of the weights and biases is called a filter. This can be represented by the following equation:

As explained in the Convolution section, the filter can take the form of an edge filter to detect edges. In a neural network, many neurons share the same filter. The weights and the filter allow the fully connected layer to act as a classifier.