FSDMeL : Few‐Shot Enabled Distributed Meta‐Learning Model for Image Recognition

ABSTRACT

Nowadays, image recognition has become a hot topic due to the widespread applicability in numerous applications, including healthcare, military applications, and infrared warnings. Numerous machine learning, deep learning, and transformer‐based image recognition techniques are designed to perform object detection. Conversely, their inability to obtain fine‐grained feature representation, increased computational complexity, and overfitting issues lower the speed and accuracy of the image recognition process. Hence, to eradicate certain downsides and to incur accurate recognition of the images, a Few‐shot enabled distributed meta‐learning (FSDMeL) framework is proposed. The proposed FSDMeL model combines a few‐shot meta‐learning strategy that primarily enables the model to learn new tasks with limited data and transfer knowledge between related classes to improve recognition performances. In line with this, the FSDMeL model also integrates various sophisticated mechanisms for feature extraction, which leverage diverse as well as relevant information to avoid overfitting issues. Moreover, the FSDMeL model also includes several benefits such as faster training times, improved generalization to unseen tasks, and the ability to provide correct detection without increasing training complexity. According to the assessment, the proposed model gains an accuracy of 98.04%, precision of 98.77%, and recall of 97.67% using the Fashion MNIST dataset compared to existing methods for image recognition.