The Tree Seed Algorithm (TSA) is a popular meta-heuristic algorithm that excels in solving optimization problems. However, TSA has some structural deficiencies and certain limitations manifested as limited population diversity, inadequate information utilization, and local optima stagnation. This paper proposes the Adaptive Tree Seed Algorithm (ATSA) to solve these mentioned shortcomings with three enhancements. First, a double-layer framework is designed to achieve a more effective balance between exploration and exploitation with the feedback mechanism. Second, based on this framework, an evolutionary classifier is designed to generate seeds intelligently for enriching population diversity. Third, a migration mechanism is proposed to avoid falling into local optima. The performance of ATSA is tested by 30 benchmark functions on IEEE CEC 2017 in comparison with 15 algorithms (TSA, STSA, TSASC, fb_TSA, EST-TSA, GWO, PSO, ABC, BOA, RSA, HBA, SMA, HHO, INFO, RUN). In addition, 5 engineering design problems are evaluated to illustrate applicability. All experimental results demonstrate that the proposed ATSA significantly outperforms other algorithms, especially in solving high-dimensional and complex problems, as verified by the Wilcoxon Signed-Rank test. The outstanding performance of ATSA makes it a promising candidate for addressing challenges in the field of swarm intelligence.