Abstract—This study proposes a new channel and spatial attention mechanism module called the Binarized Convolutional Block Attention Module (B-CBAM), which is characterized by binarized channel attention feature map, optimized activation function for the binarization, and spatial attention feature map. Unlike the conventional Convolutional Block Attention Module (CBAM), the binarized attention feature map and the best choice of the activation function for the binarization allow accelerating the learning curve and eliminating the need for Max pooling and multilayer perceptron. The binarized convolution significantly reduces the computational load required by the attention module, with memory requirements being only 1/32 of those for full-precision weights. After training for 100 epochs, the average error rates improved by 21% on CIFAR-10 and 9% on STL-10. The optimized B-CBAM model, after parameter tuning, achieves error rates very close to those of the non-quantized model, with an error rate of around 9% on CIFAR-10.
 
Keywords—binarized, residual model, attention mechanism, Convolutional Neural Network (CNN) efficiency, feature maps binarization


Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).