Deep learning approaches for reliable object detection under adverse weather

In severe weather, meteorological elements including sand, rain, fog, and snow reduce picture quality, making object recognition challenging in computer vision. In low-visibility conditions, CNN, RCNN, and faster R-CNNs cannot overcome weather distortions, reducing their performance. CNN models lack feature augmentation procedures, RCNN is computationally inefficient, and Faster R-CNN is robust but has visual degradation issues in severe instances. This paper proposed a Hybrid AOD-Net + CycleGAN + Faster R-CNN model for image augmentation, domain adaptation, and reliable object recognition to address such restrictions. The model&s;s evaluation under various adverse weather conditions yielded a Mean Average Precision (MAP) of 0.505, Precision of 0.94, and Recall of 0.92 for sandy weather, 0.511, Precision of 0.93, and Recall of 0.90 for rainy weather, 0.521, Precision of 0.92, and Recall of 0.89 for foggy weather, and 0.504, Precision of 0.93, and Recall of 0.90 These findings show that picture-based models outperorm Faster R-CNN, RCNN, and CNN models augmentation and domain adaption techniques.