<p>The formation and development of natural fractures in tight sandstone reservoirs are governed by a combination of stratigraphic structure, lithological properties, and stress conditions. These fractures often exhibit irregular geometries, significant variations in height, and complex filling materials, leading to intricate conventional logging responses with pronounced multi-solution ambiguities that complicate accurate identification. To address this challenge, this study proposes a multi-model selective coupling identification method. This approach incorporated data cleaning, augmentation, and resampling techniques during the preprocessing phase. Subsequently, multi-dimensional feature extraction and cascade-based feature selection were performed, followed by optimizing model parameters using random search, Bayesian optimization, and grid search algorithms. High-performing models were selected via an evaluation framework. These models were then coupled through voting mechanisms to construct a robust identification model capable of deeply exploring the nonlinear relationship between fractures and logging data. The proposed method achieved an 85.19% fracture identification accuracy in blind tests involving 27 fracture segments across three wells, demonstrating strong identification capability. This methodology provides a valuable reference for fracture identification in hydrocarbon reservoirs within the Hongde area.</p>