I am trying to use morlet wavelet on BCIC-IV-2a dataset to create a 2D image. Since, the dataset is 1000 samples long and with 22 channels, after applying the transform the shape becomes (22,127,1000) where 127 is of course wavelet coefficients.
My aim is to apply pre-trained Resnet model upon this image, for that I resized the image to (224,224) from (127,1000).
Can you suggest further steps that can be taken for proper training ? Or have I missed necessary processing on the wavelet image here ?
consider normalizing the wavelet coefficients and applying dimensionality reduction techniques like PCA to reduce the high dimensionality of the CWT output. Handle the 22 channels appropriately, either by concatenating them into a 3D tensor or stacking them to capture dependencies. Fine-tune the pre-trained ResNet by unfreezing certain layers with a lower learning rate and leverage transfer learning. Preserve the aspect ratio of the time-frequency representation during resizing to avoid distorting patterns.
Dear Dibbo Dey
,The processing of brain signals for Motor imagery (MI) classification to have better accuracy is a key issue in the Brain-Computer Interface (BCI). While conventional methods like Artificial neural network (ANN), Linear discernment analysis (LDA), K-Nearest Neighbor (KNN), Support vector machine (SVM), etc. have made significant progress in terms of classification accuracy, deep transfer learning-based systems have shown the potential to outperform them. BCI can play a vital role in enabling communication with the external world for persons with motor disabilities.
Article CWT Based Transfer Learning for Motor Imagery Classification...
Regards,
Shafagat
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs