Figure 3
The convolutional neural network architecture begins with stacked input images labelled Input. A highlighted region from the input connects by dotted lines to feature maps in Conv Layer 1. The feature maps then pass to Pooling Layer 1, followed by Conv Layer 2 and Pooling Layer 2. Dotted outlines show local regions transferred between successive feature maps. An ellipsis indicates additional intermediate processing before the network reaches a fully connected layer. The convolution and pooling stages are enclosed within a dashed region labelled Feature Extraction. The fully connected neural network is enclosed within a separate dashed region labelled Classification and contains several connected nodes arranged in layers, with ellipses indicating additional nodes. Four arrows lead from this network to output boxes labelled Class 1, Class 2, Class 3 and Class 4.Architecture of a convolutional neural network (CNN)