Kolmogorov–Arnold Propagation Network Model-Assisted Rapid and Precise Identification of Oxide Component States in Complex Solid Waste

Accurate and efficient identification of oxide components and their quantitative proportions serves as a fundamental prerequisite for the sustainable reutilization of solid waste materials. However, conventional characterization approaches are severely constrained by either sophisticated technical limitations or complex mixed-component scenarios, rendering it challenging to achieve low-cost and high-efficiency identification and quantification of oxide constituents within solid wastes. To address this critical bottleneck, herein we develop a novel artificial intelligence-enabled analytical strategy integrated with a Kolmogorov–Arnold network (KAN) visual model, which enables rapid identification of mixed oxide components solely relying on visual photographs of oxides and ultraviolet–visible (UV–vis) absorption spectra. Systematic validation experiments were conducted on four typical oxides, and the proposed model delivers outstanding identification performance, with overall accuracy exceeding 90% for all target oxides. Crucially, featuring a compact architecture and low computational overhead, the KAN visual model is highly amenable to deployment on edge devices, enabling on-site and real-time analyses. Furthermore, we extended this intelligent identification platform to the representative solid waste red mud. By coupling visual image features and UV–vis spectral data, the KAN visual model achieves precise discrimination of oxide components and accurate quantitative proportion analysis in red mud samples derived from diverse geographical regions, which are in excellent consistency with X-ray diffraction pattern results. This work fully demonstrates the immense application potential and superior reliability of the developed KAN visual recognition model for the efficient and portable component analysis of complex mixed oxide systems.