OExVT-D2CN: One-Shot Enabled External Vision Triplet Attention-Based Distributed Deep Learning Model for Image Recognition

Evaluation of one-shot-based image recognition contributes to deploying a single input to train a deep learning model regarding the visual concepts. Over the past decades, many researchers have expanded their research on one-shot learning-based image recognition, but failed to achieve accurate performance due to overfitting, insufficient samples, and poor interpretation. Most of the existing approaches cannot provide accurate results when adopts blur images as training samples, which limits efficiency and performance improvements over image recognition experiments. To alleviate such consequences, this research proposes a one-shot-enabled external vision triplet attention-based distributed deep convolutional neural network (OExVT-D2CN) model for effective image recognition. On top of that, this research presents an ExVT module to pay attention to a specific region of the input to enhance recognition stability as well as interpretability. The OExVT-D2CN model accelerates training efficiency by optimizing the input and attenuating challenges on complex datasets and overfitting issues. Based on these hybrid implementation techniques, the proposed model precisely recognizes the objects with minimal computation cost and memory constraints. Furthermore, the OExVT-D2CN model acquires significant results under a one-shot condition, which attains a maximum of 99.41% accuracy and 96.92% precision using the PASCAL VOC-2012 dataset when compared to other state-of-the-art methods.