Abstract: Semantic segmentation of road scenes plays a crucial role in many fields such as autonomous driving, intelligent transportation systems and urban planning. Through the precise identification and segmentation of elements such as roads, pedestrians, vehicles, and traffic signs, the system can better understand the surrounding environment and make safe and effective decisions. However, the existing semantic segmentation technology still faces many challenges in the face of complex road scenes, such as lighting changes, weather effects, different viewing angles and the existence of occlusions. Combined with the actual road scene image, this paper improves DeeplabV3+ network and applies it to semantic segmentation of road scene image, and proposes a semantic segmentation method of road scene image based on improved DeeplabV3+ network. By adding enhancement strategies for road scene images and hyperparameter adjustment, the method improves the training process of DeeplabV3+ network, and uses SK attention mechanism to improve the feature fusion module in DeeplabV3+, so as to improve the segmentation effect of road scene images. After the validation of Cityscapes and other data sets, the segmentation accuracy index mIoU of the proposed method reaches 79.8%, which can predict better semantic style effect, effectively improve the segmentation performance and accuracy of the model, and achieve better segmentation index results in the comparison network, and the subjective visual effect of the segmentation is also better.

Lihua Bi, Xiangfei Zhang, Shihao Li and Canlin Li, “A Semantic Segmentation Method for Road Scene Images Based on Improved DeeplabV3+ Network” International Journal of Advanced Computer Science and Applications(IJACSA), 15(8), 2024. http://dx.doi.org/10.14569/IJACSA.2024.0150883

@article{Bi2024,
title = {A Semantic Segmentation Method for Road Scene Images Based on Improved DeeplabV3+ Network},
journal = {International Journal of Advanced Computer Science and Applications},
doi = {10.14569/IJACSA.2024.0150883},
url = {http://dx.doi.org/10.14569/IJACSA.2024.0150883},
year = {2024},
publisher = {The Science and Information Organization},
volume = {15},
number = {8},
author = {Lihua Bi and Xiangfei Zhang and Shihao Li and Canlin Li}
}