Multi-dimensional top-k flow detection in high-speed networks faces significant challenges due to memory inefficiency and static resource allocation. Traditional approaches employ independent sketches for each dimension, resulting in substantial redundancy as flows frequently rank high across multiple dimensions simultaneously. Static memory partitioning further limits adaptability to dynamic traffic patterns, degrading detection performance under varying network traffic. We present DCT-Sketch, a unified memory architecture that exploits cross-dimensional correlations through collaborative storage and adaptive resource management. Our approach integrates a Collaborative Estimation Pool with dimension-specific tables, dynamically adjusting memory allocation based ohn observed traffic characteristics. The system employs composite scoring mechanisms to prioritize flows across multiple dimensions while eliminating redundant storage. Comprehensive evaluation on real-world traces demonstrates superior performance compared to existing approaches, achieving substantial memory savings while maintaining high detection accuracy under diverse network conditions. These results validate the effectiveness of collaborative multi-dimensional flow monitoring for resource-constrained environments.