Deep convolutional fuzzy systems (DCFS) have emerged as a dominant approach for tackling high-dimensional problems. However, their performance is compromised by exponential rule growth (e.g., 59,049 rules for 10-dimensional inputs) and numerical instability (trigger strength \(\le\) 0.00098), which hinder real-world deployment in safety-critical scenarios. Current mainstream parameter optimization methods, while significantly enhancing system performance, demand substantial memory and computational resources. Consequently, structural optimization through external system tuning has become a prevalent and efficient design method in the machine learning field. This paper proposes a structurally optimized approach for deep patch learning fuzzy systems in classification (DPFSCAs) from a topological perspective, employing a bag of tricks that enables synergistic optimization via multi-strategy integration of three novel techniques–feature combination (FC), drop connect (DC), and subsystem deletion (SD). The study explores strategies impacting DPFSCAs performance and proposes corresponding structural and performance optimization methods. Experiments on diverse datasets demonstrate that FC, DC, and SD, both individually and integrated, enhance DPFSCAs classification performance through effective synergistic effects.
