Object detection in adverse weather conditions for small object or objects present at a larger distance face safety threat in autonomous vehicles (AV). Detection speed and accuracy are the key requirements in such challenging environment. The proposed model in this paper addresses it with Small Object detection with Improved Loss-YOLOv8 (SOIL-YOLOv8) focusing on the enhanced backbone structure with SPPF module, incorporation of NAS-FPN feature pyramid, attention mechanism and dynamic head based on hybrid adaptive IoU loss. Experiments are conducted on BDD100k and CADC datasets to diversify the training and validation results for small object detection. The results exhibit that detection accuracy and speed in terms of models efficiency improves in comparison to the baseline model. Additional, performance increases significantly in comparison to state-of-the-arts models.