Sravya, S.Pavithra, U.Kevala, V.D.Lal, S.Nalini, J.Chintala, C.S.2026-02-032025Journal of the Indian Society of Remote Sensing, 2025, , , pp. -0255660Xhttps://doi.org/10.1007/s12524-025-02289-7https://idr.nitk.ac.in/handle/123456789/20654An effective algorithm for road network extraction from satellite images can lighten the challenges in numerous applications, including GPS navigation, city planning, traffic control, and map updating. This paper introduces an effective CartoRoadNet model tailored for road network extraction from high resolution satellite images. The proposed model is designed by incorporating advanced mechanisms such as Shuffle Block with Channel Weightage (SCW), Squeeze and Excitaion - ResNet (SE-ResNet) module, and Pixel Adaptive Field-of-View (PA_FoV) within a U-Net architecture. A new dataset named Cartosat-2E road is also introduced for road network extraction which consists of images captured by the Indian Remote Sensing Satellite (IRS) Cartosat-2E. The performance of the proposed CartoRoadNet model is evaluated on the new dataset and publicly available Massachusetts and CHN6-CUG road datasets. The performance of the proposed CartoRoadNet is compared with benchmark deep learning models and it demonstrates superior performance, achieving (81.45, 87.79, 0.87.02%) F1-score, (71.58, 79.97, 78.47%) mIoU and (62.90, 75.58, 74.03%) Kappa Coefficient on the Cartosat-2E, Massachusetts and CHN6-CUG road datasets respectively. These results highlight the model’s robust feature extraction and spatial attention capabilities, ensuring precise segmentation outcomes while maintaining a less number of parameters. This work advances the field of image segmentation by offering a highly accurate, efficient and generalized framework suitable for various high-precision segmentation tasks especially for dataset covering the Indian region. © Indian Society of Remote Sensing 2025.Cartosat-2E road datasetConvolutional Neural Network (CNN)Road segmentation