Enhanced Image Classification through Layer-Wise Feature Concatenation in Deep Neural Networks

Image classification plays a pivotal role across various applications in sectors such as healthcare, agriculture, security, and surveillance. However, deploying large-scale models on resource-constrained edge devices remains a significant challenge. To address this issue, this paper introduces a novel, compact image classification framework that is specifically optimized for deployment on edge devices. The proposed model combines a simplified head block with four sequential stages, designed to balance accuracy with computational efficiency effectively. Evaluations on two datasets demonstrate the model's efficacy: it achieves an accuracy of 90.7% with only 0.11M parameters on the CIFAR-10 dataset, and 94.59% accuracy with 0.79M parameters on the Butterfly and Moth dataset. These results underscore the potential of the proposed framework as a practical solution for efficient, real-time image classification in environments where computational resources are limited.