Image Classification Based on Triplet Neighborhood Attention Transformer

Over time, artificial intelligence networks have grown and flourished, with numerous innovative and practical applications becoming an integral part of daily life. Moreover, modern designs have evolved alongside the growth of AI, becoming increasingly complex. In the field of computer vision, a wide variety of neural networks now compete with one another, tackling tasks such as image classification, object detection, and segmentation, with many established use cases already in place. This study focuses on image classification and proposes the Triplet Neighborhood Attention Transformer architecture, which combines the features of Neighborhood Attention Transformer and Triplet Attention. Without significantly increasing the number of parameters, this approach enhances neural network accuracy, enabling better performance for the neural network.