Oil spills remain among the major environmental threats to marine ecosystems and coastal operations, making rapid and accurate detection essential for effective monitoring and emergency response. Synthetic aperture radar (SAR) imagery is widely used for this purpose because it enables daytime, nighttime, and all-weather conditions. Nevertheless, reliable oil spill segmentation remains challenging because SAR images are affected by speckle noise, low contrast, and visual similarities between oil-contaminated areas and surrounding sea surfaces. This paper presents a deep learning-based semantic segmentation framework using U-Net as the primary architecture for binary oil spill detection in SAR imagery with potential integration into UAV-assisted monitoring systems. The dataset comprised 1,112 Sentinel-1 SAR images acquired under diverse environmental and oceanic conditions. The workflow included SAR image preprocessing through logarithmic transformation, wavelet-based denoising, contrast enhancement using contrast-limited adaptive histogram equalization, and resizing to 256 × 256 px before training. The U-Net model was trained using the Adam optimizer, weighted sparse categorical cross-entropy loss, a batch size of 16, and 30 training epochs to improve segmentation performance under class imbalance. The results showed stable convergence, with the U-Net model achieving an accuracy of 93.33%, a mean Dice score of approximately 0.59, and a mean intersection over union above 0.52. Qualitative predictions also demonstrated accurate localization of major oil spill regions with clear boundaries and limited background noise. Benchmarking against a previously developed DeepLabV3+-based model confirmed the suitability of U-Net as a practical and computationally efficient solution for SAR-based oil spill monitoring and intelligent UAV deployment