Seagull optimization algorithm (SOA) exhibits certain weaknesses such as poor accuracy and a tendency to stagnate in local optimal solutions when solving complex optimization problems. This paper suggests an enhanced variant of SOA, referred to as planar-mirror reflection imaging learning based SOA (PRIL-SOA), to address these limitations. First, we present the novel nonlinear strategies for adjusting the employing variable A and control parameter B are presented to achieve a balance between global and local search capabilities. Second, a modified position update equation is devised that incorporates velocity components and personal history best positions, thereby enhance solution precision. Third, a new PRIL strategy is introduced to maintain diversity and prevent premature convergence. To validate the performance of PRIL-SOA, we conduct a series of benchmark tests, including 23 classical functions and a feature selection problem involving 21 datasets are used. The results indicate that PRIL-SOA consistently outperforms basic SOA and other meta-heuristics. The average search success rate of PRIL-SOA on benchmark test problems is 91.3 %, with 21 out of 23 problems achieving the theoretical optimal value. Compare with SOA, mountain gazelle optimizer (MGO), whale optimization algorithm (WOA), hunger games search (HGS), HHO-based joint opposite selection (HHO-JOS), modified SCA (MSCA), and exploration-enhanced GWO (EEGWO), the average success rates of PRIL-SOA is better to 86.95 %, 78.26 %, 82.61 %, 65.22 %, 56.52 %, 60.87 %, and 4.35 %, respectively.