Abstract: Nitrogen and phosphorus in farmland soils are essential nutrients for crop growth. Their imbalance not only reduces crop yield but also exacerbates agricultural non-point source pollution, posing potential risks to regional ecological security. Traditional manual sampling monitoring is limited by low efficiency, high cost, and restricted spatial coverage, making large-scale dynamic monitoring difficult. In contrast, multi-source remote sensing offers an efficient and feasible approach for monitoring soil nitrogen and phosphorus due to its broad-scale, rapid, and non-destructive advantages. This review systematically summarizes domestic and international research. First, the main remote sensing data sources, including satellites, hyperspectral drones, and ground-based spectroscopy, and their applicable scenarios are outlined. Second, the review compares and analyzes the main inversion approaches, including direct inversion based on spectral features, indirect inversion based on vegetation characteristics, and integrated evaluation through ensemble decision-making, highlighting their respective strengths and limitations. Furthermore, the potential and innovative applications of deep learning models in high-dimensional nonlinear feature extraction, spatio-temporal dynamic modeling, and self-attention mechanisms are discussed. The review emphasizes that multi-source data fusion, incorporation of physical mechanism constraints, and lightweight, interpretable deep learning models will be key directions to enhance soil nitrogen and phosphorus monitoring accuracy and enable intelligent management. This work provides a theoretical reference for the development of remote sensing inversion technologies for soil nitrogen and phosphorus and offers technical support for precision agriculture and environmental governance.