Hybrid Convolutional Neural Network-Transformer Based Edge Artificial Intelligence for Wireless Image Recognition

From the past few years, wireless image recognition has become increasingly important in applications like UAV surveillance, Internet of Things (IoT) monitoring and smart camera networks. Traditional models had faced challenges with sensitivity to wireless transmission noise, inability to capture both local and global features effectively. To overcome these challenges, this research proposes Hybrid Convolutional Neural Network-Transformer Based Edge Artificial Intelligence (HCNN-Edge AI) framework for wireless image recognition. Initially, images are collected from diverse wireless image dataset, followed by preprocessing for removing the noise, normalize and augment the data for improved generalization. Then, feature extraction is done by using hybrid model where CNN model captures local spatial patterns and Vision Transformer (ViT) models the global dependencies. After that, the embeddings are processed by lightweight MultiLayer Perceptron (MLP) which are deployed on the edge devices. This approach ensures robust recognition under different wireless conditions by maintaining efficiency which is suitable for resource-constrained edge environments. Finally, recognized outputs are transmitted back to central server or application layer for decision-making tasks like monitoring. The proposed HCNN-Edge AI achieved better results in terms of top-1 accuracy (90%), top-5 accuracy (97%) and recognition accuracy (96%) when compared with existing models.