Face recognition (FR) has become an essential biometric technology for security, surveillance, and identity verification in modern intelligent systems. However, existing FR approaches often suffer from reduced accuracy under challenging conditions like pose variation, illumination changes, occlusion, and background noise. To address these limitations, a novel FA-FAS Net is proposed for robust and accurate face recognition. the input facial image is obtained from a face image dataset and passed through a face detection and cropping stage to isolate the facial region from the background. The detected face fed into preprocessing, it includes resizing to a standard dimension, normalization to maintain consistent pixel intensity distribution, and denoising to remove unwanted noise and enhance image quality. The pre-processed image is subsequently forwarded to the feature extraction module, where a deep convolutional neural network based on RegNet is employed to learn discriminative facial representations and generate robust feature maps. These extracted features are then utilized in the detection and recognition stage, where a backbone network inspired by VGG16 supports the object detection framework implemented using Faster R-CNN, which accurately localizes and classifies the detected face. Finally, the system outputs the recognized identity, demonstrating the efficiency of the integrated deep learning framework for reliable and high-precision FR in real-world applications. The FA-FAS Net maintains high accuracy levels of 98.87% based on the gathered dataset. The FA-FAS Net enhances the total accuracy by 0.99%, 4.07% and 4.92% better than FER, CNN, and Deep neural network respectively.