CNN+BLSTM network

I'm working on CNN+BLSTM project on matlab2019 that provides the training of these 2 blocks together. Which layer can I use to make the CNN output (feature maps) compatible with BLSTM input? I want to convert each feature map into a vector. Flatten layer doesn't solve my issue.

Answer 1

[praguna manvi]

Hi @zeinab al hawraa koumayha,

To integrate "CNN" and "LSTM" blocks together, you should apply the "CNN" layer to each input in the sequence. The extracted feature maps can then be passed onto "BLSTM" or "LSTM" layers for sequential processing. Here's an existing example in R2024b that demonstrates this:

openExample("deeplearning_shared/SequenceClassificationUsingCNNLSTMNetworkExample")

For more information, check out the "Define 2-D CNN LSTM Architecture" section at the following link:

https://www.mathworks.com/help/deeplearning/ug/sequence-classification-using-cnn-lstm-network.html/