Optimized Culprit Identification Using MobileNet and Attention Mechanisms
Researchers Savitha N J and Lata B T have proposed a new deep learning framework for automated culprit identification in surveillance systems, detailed in a paper submitted to arXiv. The model integrates a lightweight MobileNet architecture with channel and spatial attention mechanisms to enhance feature representation by focusing on discriminative regions while suppressing background noise. Optimized using the Adam Optimizer, the framework was tested on benchmark datasets including LFW, CASIA-WebFace, and VGGFace2 under realistic conditions involving variations in illumination, pose, and occlusion. The study reports a high classification accuracy of 97.8%, outperforming conventional models such as baseline CNN, ResNet, and standard MobileNet. Confusion matrix analysis and ROC-AUC evaluations confirm robust performance with minimal misclassification. Crucially, the approach maintains low computational complexity and reduced inference time, making it highly suitable for real-time surveillance and edge-based applications where computational resources are limited. This advancement addresses the critical need for both high accuracy and efficiency in modern security infrastructure.
Wire timeline
Optimized Culprit Identification Using MobileNet and Attention Mechanisms
Researchers Savitha N J and Lata B T have proposed a new deep learning framework for automated culprit identification in surveillance systems, detailed in a paper submitted to arXiv. The model integrates a lightweight MobileNet architecture with channel and spatial attention mechanisms to enhance feature representation by focusing on discriminative regions while suppressing background noise. Optimized using the Adam Optimizer, the framework was tested on benchmark datasets including LFW, CASIA-WebFace, and VGGFace2 under realistic conditions involving variations in illumination, pose, and occlusion. The study reports a high classification accuracy of 97.8%, outperforming conventional models such as baseline CNN, ResNet, and standard MobileNet. Confusion matrix analysis and ROC-AUC evaluations confirm robust performance with minimal misclassification. Crucially, the approach maintains low computational complexity and reduced inference time, making it highly suitable for real-time surveillance and edge-based applications where computational resources are limited. This advancement addresses the critical need for both high accuracy and efficiency in modern security infrastructure.
cs.AI updates on arXiv.org