This research proposes a transfer learning approach using MobileNetV1 and MobileNetV2 to detect cataracts and glaucoma from eye images. The models are optimized with depth-wise separable convolutions to build lightweight deep neural networks. The study uses publicly available datasets containing images of cataract, glaucoma, and normal eyes. The proposed models achieve high accuracy in classifying these conditions, outperforming other models such as VeggNet, ResNet, and traditional deep learning architectures. The models are evaluated using performance metrics like accuracy, precision, recall, and F1-score. The results show that MobileNetV1 and MobileNetV2 achieve high accuracy in both binary and multi-label classification tasks. The study also highlights the efficiency of MobileNet models for deployment on resource-constrained devices. The proposed method includes three dense layers and a global pooling layer to enhance model performance. The models are tested on various datasets, including a new benchmark dataset, and demonstrate superior accuracy in detecting eye diseases. The study concludes that the proposed models are effective for early and accurate detection of cataracts and glaucoma, with high accuracy and efficiency. Future work includes expanding the dataset to include more eye conditions and improving real-time image processing capabilities for clinical applications.This research proposes a transfer learning approach using MobileNetV1 and MobileNetV2 to detect cataracts and glaucoma from eye images. The models are optimized with depth-wise separable convolutions to build lightweight deep neural networks. The study uses publicly available datasets containing images of cataract, glaucoma, and normal eyes. The proposed models achieve high accuracy in classifying these conditions, outperforming other models such as VeggNet, ResNet, and traditional deep learning architectures. The models are evaluated using performance metrics like accuracy, precision, recall, and F1-score. The results show that MobileNetV1 and MobileNetV2 achieve high accuracy in both binary and multi-label classification tasks. The study also highlights the efficiency of MobileNet models for deployment on resource-constrained devices. The proposed method includes three dense layers and a global pooling layer to enhance model performance. The models are tested on various datasets, including a new benchmark dataset, and demonstrate superior accuracy in detecting eye diseases. The study concludes that the proposed models are effective for early and accurate detection of cataracts and glaucoma, with high accuracy and efficiency. Future work includes expanding the dataset to include more eye conditions and improving real-time image processing capabilities for clinical applications.