Image recognition based on deep residual shrinkage Network

• Published in: 2021 International Conference on Artificial Intelligence and Electromechanical Automation (AIEA) pp. 334 - 337
• Main Authors: Li, Yu, Chen, Hongguan
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.05.2021
• Subjects: attention mechanism; component; Data models; Deep learning; deep residual shrinkage network; Feature extraction; Image recognition; soft threshold function; Supervised learning; Training; Transfer learning
• DOI: 10.1109/AIEA53260.2021.00077

Deep residual shrinkage network is a novel and improved algorithm based on deep residual shrinkage network by introducing attention mechanism and soft threshold function. The core of the deep residual shrinkage network can be expressed as: when processing the image sample data, the important information features are extracted by the attention mechanism, and then the negligible information features are eliminated by the soft threshold function, only important information is retained, and these important information features are processed to get the final results, so as to improve the accuracy. The deep residual shrinkage network was applied to MNIST dataset and CIFAR-10 dataset to determine the feasibility of this method in image recognition, and then the network was applied to the facial expression dataset made by the author based on FER2013. Experimental results show that this method can effectively improve the accuracy of image recognition.