Early identification of dyslexia is important for appropriate intervention for children at risk of writing problems. In this paper we illustrate a novel image data augmentation technique integrated with edge computing architecture for deep learning assisted automatic dyslexia diagnosis. The model utilizes pre-trained VGG16 model, deployed on edge server framework to be fine-tuned on classifying the handwriting samples into dyslexic or non dyslexic. For preventing overfitting with a small dataset, we applied multiple augmentation methods such as rotation, zoom, flip and shift which enriched the training body set. Experimental results indicate that the method of data augmentation substantially improved the performance of the model, resulting in an accuracy rise from 90% to 95% and substantial increase in precision, recall and FI-score. The findings suggest that the combination of data augmentation, transfer learning and edge computing is a robust approach for dyslexia screening. This method could be used to help educators and clinicians in early screening and custom education aids for children.