Abstract: Objective This study aims to address the monitoring challenges in femtosecond laser welding of glass and copper dissimilar materials, which arises from small heat-affected zone and ultrashort processing time. Methods Acoustic emission signals were utilized to monitor femtosecond laser welding process of glass and copper. A feature termed relative “root mean square sum” was proposed to correlate formation of welded connections with acoustic emission signals. Furthermore, a neural network model was introduced to enable intelligent judgment of welding state. Performance of convolutional neural network was enhanced by increasing convolutional kernel size, integrating multi-dimensional features and incorporating prior knowledge through a customized loss function. Results The optimized model improved accuracy of successful weld detection from 89% to 96%, enabling high-precision monitoring of welding process. Conclusion The proposed method, based on acoustic emission signals and neural networks, offers a novel approach for monitoring the femtosecond laser welding of glass-metal dissimilar materials, significantly improving accuracy and intelligence of welding quality assessment.