Multi-label spacecraft electrical signal classification method based on DBN and random forest
Fig 3
The topology of the SAE network.
The representation ability of only auto-encoder is too limited. Thus the stacked auto-encoder composed of multiple auto-encoder is able to greatly improve the representational power. The activation value of one auto-encoder is the input of the upper auto-encoder. It can be used as a pre-training technique and it should belong to the unsupervised learning method for it doesn’t need label information at all. The topological graph of the SAE is shown in Fig 3.