The quantitative analysis of the effective atomic number in X-ray projection imaging is constrained by beam hardening effects and nonlinear spectral distortion induced by detector response. To address these limitations, this paper proposes a composite correction method for thickness-decoupled effective atomic number inversion. The method establishes a folded energy spectrum model integrating the attenuation characteristics of a composite X-ray source with detector response matrices, and a response-corrected spectral database for materials with different atomic numbers is established through Monte Carlo simulation. The Spectral Mass-Attenuation Linearisation (SMAL) method is employed to map nonlinear attenuation curves onto a unified linear domain, followed by weighted least-squares spectral matching to perform effective atomic number () inversion. Experimental results obtained using standard materials demonstrate that the proposed method achieves high quantitative accuracy in effective atomic number estimation, with the inversion error controlled within ± 0.2. In defect detection applications for carbon fibre composites, the method clearly distinguishes samples with similar grey levels but different compositions, highlighting its practical value in non-destructive testing.