This study addresses the suboptimal efficiency of regional science and technology innovation resource allocation by proposing a multi-source data fusion algorithm grounded in set pair analysis. Through a systematic review of theories and methodologies on science and technology innovation resource allocation, we design a theoretical framework integrating three modes: single-driven, joint-driven, and collaborative resource optimization. The proposed algorithm extracts opposition, uniformity, and difference degrees from sensor data using set pair analysis, constructs a connection matrix, and employs a signal-to-noise ratio weighting mechanism for weighted fusion. Simulation experiments demonstrate the algorithm’s superior accuracy and stability, with absolute errors reduced by 30–50% compared to traditional methods. An improved DEA model evaluates regional resource allocation efficiency, revealing nonlinear input-output relationships and Pareto optimization trends across 15 Chinese provinces. Results indicate that optimized resource allocation enhances multi-source data fusion capabilities, accelerates convergence by 37%, and improves regional innovation competitiveness. This work provides actionable insights for policymakers to harmonize government-market dynamics and foster sustainable innovation ecosystems.