Abstract: In order to quickly and accurately quantitatively evaluate the metal spatter to optimize the process and ensure the welding quality. This study focuses on the variable beam profile (VBP) laser welding process of 1060 aluminum alloy and develops an in-situ keyhole depth measurement system based on optical coherence tomography (OCT). An innovative 1DCNN-BiLSTM deep learning composite model is proposed, leveraging the distinct characteristics of the two network units to perform local-global temporal feature extraction, achieving quantitative evaluation of spatter status. Results indicate that the constructed model achieves 99.69% accuracy in identifying spatter status, providing guidance and closed-loop feedback for optimizing the VBP laser welding process and quality control.