Improving Image Recognition Performance through an Improved Dual Convolutional Neural Network with Recurrent Integration and Residual Model
Assist. Prof. Dr. Ahmed Abdullah Ahmed
Mr. Kanar Tariq
Abstract:
This observe proposes a new Convolutional Neural Network (CNN) framework for image category, the usage of both CNN and Recurrent Neural Networks (RNNs) for superior characteristic mastering. The technique combines RNNs in CNNs to offer neighborhood and temporal correlations can be extracted. Additionally, a new "ShortCut3-ResNet" module, triggered by the ultimate ResNet connections, facilitates easy float of data over layers. Moreover, the twin optimization model optimizes cooperatively on the convolutional and fully related tiers. This correction is evaluated the use of the CIFAR-10 statistics set. Experiments display the efficiency of the proposed method, achieving better overall performance in comparison to existing methods in phrases of accuracy and sample length The study also investigates the effect of activation characteristic, sampling strategies, pooling strategies, and dual optimization so, and provides precious insights for optimizing CNN overall performance.
