<p>Adverse weather and bad lighting in real-world scenarios cause performance degradation of object detection models. Currently, many solutions proposed to address this problem concentrate only on a certain kind of adverse weather. The coverage of adverse lighting scenes is not comprehensive enough, thus making it difficult for application to other scenes. Aiming to develop an algorithm that is capable of applying to vehicle detection under various adverse lighting conditions (rainy, foggy, snowy, and dark night), this paper proposes an INCGM-YOLO algorithm based on YOLOv7. InceptionNeXt block is utilized in the backbone network to optimize the backbone structure, and replace part of the ELAN structure in the head with C3 module. In addition, the global attention mechanism is added at the end of the backbone, MP and ELAN structures with GAM attention are designed in the head, which enhances the attention to effective features. Finally, the original loss function is replaced with MPDIoU to improve the accuracy of the bounding box regression. A series of experiments on the adverse weather and dark night datasets show that INCGM-YOLO reaches 76.59% and 51.68% for mAP0.5 and mAP0.5:0.95, respectively, which is improved over baseline YOLOv7 by 7.09% and 5.39%. Comparing with other state-of-the-art algorithms for vehicle detection under adverse lighting conditions, our model achieves the best performance on mAP0.5, mAP0.5:0.95 and F1 score. Visualization of detection results proves that the proposed INCGM-YOLO can significantly improve the detection precision and effectively realize vehicle detection under adverse lighting conditions.</p>