Dual-stream image recognition network based on visual field-of-view mechanism

It is significant that the field-of-view mechanism plays a role in the visual system of the primate, such as reducing information overload and focusing on important information, which can ensure optimal information processing capabilities and an allocation mechanism for computing resources is implemented. Notably, this mechanism operates prior to information calculation, distinguishing it from the attention mechanism, which influences the calculation process itself. We utilize the mechanism in a dual-stream network for image classification for the first time and apply an ideology of reinforcement learning to improve the accuracy of prediction. More importantly, our experiments certify the performance of our method in the Mini-ImageNet and CUB-200-2011 dataset.