Pathologic myopia (PM) is a severe form of myopia that, if not treated early, can lead to permanent vision loss. However, diagnosing PM from retinal fundus images is often slow, subjective, and limited by access to specialists. To address this, we present a dual deep learning system that combines a CBAM-enhanced GM-U-Net for pinpointing lesions and a custom lightweight classifier, PMNet, for identifying PM cases. Our approach also uses focal loss and saliency maps to handle data imbalance and enhance interpretability. Trained on a curated mix of three public datasets, the model outperforms existing methods in both lesion segmentation and disease classification, with an overall accuracy of 96.11%, an AUC-ROC of 0.9938, and an MCC of 0.89. This work highlights the potential of AI to support faster, more accessible PM diagnosis in clinics and remote settings.