<p>Cervical cancer is a very preventable but fatal cancer in women all over the world, primarily because of time delays in early diagnosis and the availability of accurate screening modalities in resource-poor settings. Deep learning has been a recent hope for developing algorithms for automating the detection of cervical cancer using medical imaging; however, most existing methods are hampered by poor feature extraction, poor boundary definition in lesion segmentation, and limited generalization in classification problems. Classic architectures, such as U-Net and VGG-16, while basic, often fail to differentiate the intricate patterns and heterogeneity of cervical lesions due to architectural constraints, including shallow depth, the absence of advanced skip connections, and poor feature reutilization. This paper suggests a deep learning-based model combining state-of-the-art image segmentation and classification methods for computer-aided detection of cervical cancer. UNet + + with nested skip connections is applied for accurate lesion segmentation, with a Dice score of 0.83 and IoU of 0.73, better than U-Net, DeepLabV3+, and Double U-Net baselines. The segmented outputs are then classified further with DenseNet-121 and Squeeze-and-Excitation blocks, which have a diagnostic accuracy of 93.48%, precision of 95.12%, recall of 93.98%, anSd AUC of 0.93. The two-stage pipeline improves boundary definition, feature extraction, and generalization over current models. The results prove the clinical usability of deep learning towards accurate, swift, and automated cervical cancer screening. Future research will integrate multi-modal data and model explainability to enhance real-world deployment on clinical and mobile health platforms.</p>