<p>The ancient Egyptian civilization employed Egyptian hieroglyphs as the formal writing system for expressing the Egyptian language. Recognizing the different characters in the script poses a challenge due to their lack of standardized representation. This variability makes understanding them quite challenging for humans without any knowledge of the script. Traditionally, researchers have employed Convolution Neural Networks (CNN) models for classifying letters in writing systems which gave us an accuracy of 86%. However, employing a deep learning approach enables the development of a recognition system designed to recognize hieroglyph script. This paper introduces a novel method for hieroglyph classification, utilizing a dataset comprising 4,210 images across 135 different classes. The proposed model is an improvement over traditional CNN systems which utilizes Siamese Network in tandem with triplet loss, minimizing intra-class variance and maximizing inter-class separation while also enabling better generalization and transferability across the dataset and unseen classes as well as integrating a Squeeze-and-Excitation (SE) attention mechanism. This amalgamation enhances feature extraction by dynamically recalibrating feature maps, accentuating informative hieroglyph features while mitigating less relevant ones. Consequently, it improves the discriminative capability of the model for accurate hieroglyph recognition and optimizes the assignment of weights between channels, resulting in superior performance compared to traditional CNN approaches as the model has achieved 90.08% accuracy. The outlined development strategy, the conducted experiments, and their outcomes are all covered in detail in this paper.</p>