A Novel Deep Learning Approach for Recognizing Stereotypical Motor Movements within and across Subjects on the Autism Spectrum Disorder

Figure 2

Illustration of the overall frequency-domain CNN architecture. Symbols “”, “”, and “” in the parentheses of the layer tags refer to convolution, subsampling, and unification operations, respectively. Numbers before and after “@” refer to the feature map size. Note that ReLU layers located after C1, C2, and F1 are not shown due to the limitation of space.