Autonomous experiments and multi-objective optimization are pivotal in advancing materials science, enabling the efficient discovery and characterization of novel materials. In multi-objective optimization for transparent conducting oxides, accurately determining the optical properties and thickness of these multilayer thin films is essential. However, this task is challenging due to the ill-posed nature of the fitting problem, which can lead to entrapment in local minima and slow processing times. Thus, precise initial guesses are necessary for effective fitting. This study addresses these challenges using Bi2O3 thin films as a case study, introducing a method that employs Convolutional Neural Networks (CNNs) to extract these properties from reflectance data of films processed by lateral gradient laser spike annealing (lg-LSA). A simulated reflectance dataset, generated using the Tauc-Lorentz (TL) model and Transfer Matrix Method (TMM), was used to train the CNN to predict TL parameters and thickness. The CNN model achieved a high mean R^2 score of 0.9754 on simulated data. The model's accuracy improved on experimental data from a mean R^2 score of 0.5892 to 0.8408 after subsequent trust region reflective (TRF) fitting. The CNN model demonstrated robustness across the lg-LSA processed stripe and accurately predicted the thickness profile, aligning well with experimental observations. This research underscores the potential of integrating machine learning techniques for high-throughput, autonomous material characterization, providing a solid foundation for future enhancements and applications across diverse material systems.