MobileNet and Attention Mechanisms Improve Surveillance Identification

A novel deep learning framework designed for automatic suspect detection in surveillance systems integrates a compact MobileNet architecture alongside channel and spatial attention techniques. This model boosts feature representation by emphasizing key areas while minimizing distracting background elements, thereby enhancing identification accuracy. It employs efficient preprocessing, attention-driven feature enhancement, and a strong classification approach refined using the Adam Optimizer. Tests were performed on established face recognition datasets such as Labelled Faces in the Wild (LFW), CASIA-WebFace, and a portion of VGGFace2, simulating realistic scenarios with changes in lighting, pose, and occlusion.