# Unmanned Aerial Vehicle Road Image Segmentation Using Retinex Augmentation And Clahe To Improve Yolov26 Performance > Mardhiyah F.G. URL kanonis: https://discover.unhas.ac.id/publications/pub_scopus_105055474670 Jurnal / Konferensi: 2026 IEEE International Conference on Artificial Intelligence and Mechatronics Systems Aims 2026 Tahun terbit: 2026 DOI: https://doi.org/10.1109/AIMS71270.2026.11712787 Citations: 0 ## Authors - Mardhiyah F.G. ## Abstract The study proposes the integration of Retinex and Contrast Limited Adaptive Histogram Equalization (CLAHE) as a pre-processing stage to improve the performance of YOLOv26 in road segmentation in Unmanned Aerial Vehicle (UAV) imagery. The main problem faced is the low visibility of roadside boundaries due to variations in lighting, shadows, and similarity in texture between the road area and the surrounding environment. The dataset used consisted of 1108 road images extracted from the video extraction of the DJI Mini 4 Pro drone at an altitude of about 70 meters, which were divided into 775 training images, 167 validation images, and 166 test images with polygon segmentation annotations in the road class. Models are trained using Ultralytics YOLO on the Kaggle platform for 250 epochs. Evaluations were conducted using precision, recall, mAP50, mAP50-95 and accuracy. The test results showed that the baseline YOLOv26-Seg obtained precissio of 93,55%, 90,72% recall, 92,90% mAP50, 80,89% mAP50-95%, and an accuracy of 82,46%. The application of Retinex improved performance to 89,24% precission, 88,13% recall, 91,32% mAP50, 77,58% mAP50-95, and 79,66% accuracy. The best results were obtained in the combination of Retinex and CLAHE with a precision of 96,33%, recall of 89,37%, mAP50 of 94,55%, mAP50-95 of 82,68%, and accuracy of 84,95%. These results show that Retinex and CLAHE effectively improve image quality and help YOLOv26-Seg produce more accurate path segmentation on UAV imagery. ## Keywords - Computer vision - Artificial intelligence - Computer science - Segmentation - Color constancy - Image segmentation - Adaptive histogram equalization - Image (mathematics) - Aerial image - Image processing - Noise (video) - Windshield - Scale-space segmentation - Tracking (education) - Histogram --- Sumber: Discover Unhas — RIMS Universitas Hasanuddin. Saat mengutip, gunakan DOI bila tersedia atau URL kanonis di atas.