Anemia caused by hemoglobin deficiency affects billions globally. Traditional detection methods are invasive, costly, and risk blood-borne infections, increasing patient morbidity and mortality. To address this, researchers have developed smartphone-based, non-invasive anemia detection systems using images of eyes, nails, or skin. These systems rely on deep learning (DL) algorithms for high accuracy, essential in medical diagnostics. However, most DL models are too computationally heavy for mobile devices. MobileNet, a lightweight DL model, suits embedded systems but may trade off accuracy compared to larger models like DenseNet121. This paper introduces SE-MobileNet, an enhanced MobileNet architecture with Squeeze-and-Excitation (SE) blocks to improve classification. On the CP-Anemic dataset, SE-MobileNet reduces parameters by up to \(50\%\) versus DenseNet121, ResNet50, VGG16, and ViT, while improving accuracy. It also boosts MobileNet’s performance.