Traditional security methods like keys, PINs, and passwords have drawbacks such as forgetting, loss, and attack vulnerability. Biometric authentication has evolved as a more secure option, utilizing advances in digital signal processing. Deep learning improves liver disease diagnosis by analyzing MRI and CT scans, resulting in greater accuracy and efficiency than standard physical evaluations. Similarly, rice crop health is critical to global food security, but traditional disease detection methods, such as visual inspections, are time-consuming, unreliable, and inadequate for large-scale cultivation. This research introduces MultiBioMedAgNet, a novel deep learning framework integrating biometric authentication, liver disease detection, and rice plant disease diagnosis. The proposed architecture employs a custom activation function to enhance feature extraction, nonlinearity, compression, and encryption, ensuring secure and efficient processing of medical and agricultural data. It is benchmarked against existing models like CNN, ResNet, DenseNet, and UNet, demonstrating superior performance in terms of accuracy and computational efficiency. The model’s effectiveness is evaluated using PSNR, SSIM, MSE, MAE, RMSE, and NRMSE, highlighting its potential for real-world applications.