• KOPPARA DILIP KUMAR

Enhancing low- lighting images (LLIE) is important in a broad spectrum of applications - in surveillance, intelligent vehicles, cell phone photography and medical imaging. Deep learning algorithms, in particular, convolutional neural networks and generative adversarial networks have shown significant performance in the enhancement of visual quality and the reconstruction of the details in dark images. Nonetheless, their use in real-time on resource-constrained devices (e.g. drones, smartphones, embedded cameras) is still a major challenge since their requirements are high in terms of computational and memory usage. In this paper, we introduce a new efficient deep learning architecture that is able to balance the performance of enhancement with low inference latency and energy use. Our approach combines the lightweight backbone networks, quantization-sensitive training and a lightweight attention-based fusion block to dynamically regulate the enhancement intensity based on local scene attributes. The proposed model is tested on various standard low-light tests and is compared against the state-of-the-art full-size and lightweight models. Our experimental findings demonstrate that our model can attain the competitive visual quality (quantified by SSIM, PSNR), and can be reduced in model size by a factor of up to 80, as well as inference speed on mobile GPUs can be sped up by 3-5x. Among them, the contributions are (i) a hybrid network design that is optimized in terms of speed and quality, (ii) dynamic fusion of multi-scale features through lightweight attention, and (iii) an end-to-end quantization-aware training strategy that is optimized to LLIE tasks. The piece of work presents real-time image enhancement on a limited hardware and introduces the opportunity of application of deep-learning enabled LLIE in mobile and embedded devices.

Intelligent Vehicles, Cell Phone Photography, Medical Imaging, Deep Learning, Low-Light Tests, Lightweight Attention Fusion.

"Efficient Deep Learning for Real-Time Low-Light Image Enhancement on Resource-Constrained Devices", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.12, Issue 10, page no.a476-a481, October-2025, Available :http://www.jetir.org/papers/JETIR2510059.pdf

"Efficient Deep Learning for Real-Time Low-Light Image Enhancement on Resource-Constrained Devices", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.12, Issue 10, page no. ppa476-a481, October-2025, Available at : http://www.jetir.org/papers/JETIR2510059.pdf