Embedded deep clustering algorithm based on improved CAE
In order to address the issue of feature loss in the network structure of the current deep convolutional embedded clustering(DCEC)algorithm,a novel network structure based on deep convolutional noise reduction autoencoder is proposed.In the new network structure,optimizing the convolutional kernel step size parameter in DCEC and adding pooling layers to enhance feature extraction while reducing network parameters and preventing overfitting phenomenon.Additionally,an upsampling layer is added to the decoder to recover the feature loss caused by pooling operation in the encoder.To further improve feature extraction effectiveness,a fully connected layer transition is introduced after the coding process.Consequently,we propose DCEC_ICAE algo-rithm which is evaluated using three classical image datasets and two classical clustering evaluation indicators.Experimental results demonstrate that our proposed DCEC_ICAE algorithm outperforms comparison algorithms in terms of clustering perfor-mance,thus validating and justifying the efficacy of our new network structure.
deep clusteringconvolutional denoising autoencoderdeep embedded clusteringneural network architecture
万方数据
维普
