Fusion-Based Digital Twin Framework with Convolutional Neural Networks (CNN) for Real-Time Threat Detection and Image Enhancement in Low-Light IoT Environments

Image acquisition in low-light environments often suffers from poor contrast, high noise, and loss of fine details, making effective enhancement challenging. Conventional enhancement methods frequently fail to balance detail preservation and noise suppression. This paper introduces a novel fusion-based technique for enhancing images captured in low-light settings. The proposed method integrates multi-scale decomposition, adaptive histogram equalization, and wavelet transform to improve image quality. Multi-scale decomposition captures critical structural information, adaptive histogram equalization adjusts intensity levels for enhanced contrast, and wavelet transform reduces noise while refining image details. Evaluation on standard low-light image datasets demonstrates that this method outperforms current state-of-the-art techniques. The enhanced images show a peak signal-to-noise ratio (PSNR) of 32.45 dB and a structural similarity index (SSIM) of 0.87, reflecting an 18% improvement in PSNR and a 12% gain in SSIM over traditional methods. Additionally, the technique reduces noise variance by 20% and enhances visual clarity, making it suitable for applications such as surveillance, medical imaging, and autonomous driving. In conclusion, this fusion-based approach provides an effective solution for low-light image enhancement, maintaining critical details and achieving improved visual quality through balanced contrast enhancement and noise reduction.