Abstract
Deep Learning based models have revolutionized EEG decoding attaining better performance than techniques using handcrafted features. Decoding and recognizing motor imagery signals accurately has always been a challenging task as these have been used in BCI for various critical applications like assisting stroke patients, controlling robotic arms, etc. This study proposes attention based CNN model consisting of an attention module having filters of various sizes that can extract features based on their importance from the motor imagery data. The proposed attention based CNN model produces good accuracy for the BCI IV 2a motor imagery dataset and the high gamma dataset.