Deep Learning methodologies have been shown to assist vegetation plantations and their stakeholders. Extraction of horticultural plantations from satellite imagery is an important example for agricultural monitoring, food security, and sustainable land management. Traditional techniques often fall short in generalizing across seasons or heterogeneous landscapes, necessitating more adaptive approaches. To address this, we utilized Conditional Generative Adversarial Networks to segment sample plantations of pomegranate and date palm in the arid regions of Rajasthan, India (Jaisalmer and Barmer districts), using Very High-Resolution Satellite (VHRS) imagery from CARTOSAT-2E. The framework, inspired by the existing Pix2Pix image translation model by Philip Isola, incorporated a U-Net generator and Patch-GAN discriminator for segmentation, trained on over 13,000 annotated image–mask pairs. Using loss functions, generator skip connections, and various activation functions, the Pix2Pix U-Net cGAN model demonstrated strong segmentation capabilities. Dice coefficient of 99.49%, IoU of 98.99%, pixel accuracy of 99.28%, precision of 99.66%, and recall of 99.32%. Focal and edge-aware loss functions further enhanced class differentiation, yielding a Dice coefficient of 94.35%, an IoU of 91.52%, a pixel accuracy of 97.95%, a precision of 97.06%, and a recall of 94.09%. While these metrics may have shown a strong performance, the true effectiveness of this approach was demonstrated through accurate vegetation classification and recreating the vegetation masks for extraction and segmentation. The workflow automates plantation detection from VHRS data. It establishes a foundation for future opportunities in horticulture segmentation and analysis, as well as its potential integration into decision-support systems for sustainable resource management.