Objective: The traditional substrate had the problems of insufficient uniformity and difficult to adapt to the detection of pesticide residues on irregular fruit surface. This study was aimed to develop a new surface-enhanced Raman spectroscopy (SERS) detection method to improve the sensitivity and applicability of thiram residue detection on fruit surface. Methods: A gold nanoparticle (AuNPs)-sodium alginate (SA) composite substrate was fabricated via a gluconolactone (GDL)-calcium carbonate (CaCO₃) slow-release crosslinking system to obtain controllable substrate shaping and improved gel homogeneity. The effects of CaCO₃ loading (w/v) and AuNPs concentrations on the performance of the substrates were optimized. Their performance was evaluated in terms of sensitivity, reproducibility, and uniformity, followed by application to the detection of thiram residues on apple peel samples. Results: When the CaCO₃ loading was 0.3% (w/v) and the AuNPs solution was concentrated 2×, the substrate exhibited the strongest Raman response. A good linear relationship (R²=0.987) was obtained between SERS signal intensity and thiram concentrations in the range of 0.005~1 mg/L. The limit of detection (LOD) for thiram residues on apple peels was reached 0.0625 μg/cm², with a recovery rate of 92.45% and a relative standard deviation (RSD) of 2.87%, meeting the requirements of the national standards. Moreover, the substrate was shown excellent uniformity (RSD=3.3%) and reproducibility (RSD=2.9%). Conclusion: The study proved that balancing gel mechanical properties with localized electromagnetic field enhancement density through slow-release crosslinking was the key to improve the detection efficiency on complex matrices, which would provide a regulatory principle for the design of flexible SERS substrates.