In this paper, we propose a hybrid convolutional neural network (CNN) architecture that can identify the melt pool width (MPW) and melt pool depth (MPD) as a continuous value based on 2-dimension melt pool images in powder bed fusion (PBF)_additive manufacturing. The melt pool images and their corresponding geometric parameters were collected from finite element methods, and linear distribution smoothing technique was implemented to combat the imbalanced dataset. The coefficient of correlation between the model prediction value and the true label was 0.988 for MPD and 0.982 for MPW, demonstrating the effectiveness of the proposed architecture. Moreover, the parameters are found within approximately 18.6 ms per image, showing that it can be used in a real-time in situ PBF process. The model was successfully integrated with a closed-loop control framework, paving the road to control PBF process parameters and ensure manufacturing quality based on both MPW and MPD, a parameter that is strenuous to extract with traditional techniques.