With more and more weapon systems being equipped in important areas such as strategic locations and docks, the problem of dynamic firing boundary estimation of weapon systems has arisen due to the random mooring of vessels in harbor basin. Compared to the complexity and time-consuming nature of real-time measurement of weapon system firing boundary using instruments, a dynamic firing boundary estimation method was proposed based on deep regression models. Initially, four typical deep regression models were constructed for dynamic firing boundary estimation, leveraging several classic deep learning architectures such as GoogleNet, ResNet, SqueezeNet, and ResNext. Through an in-depth comparative analysis of the strengths and weaknesses of each model's predictive results, a novel dynamic feature-weighted fusion model was introduced. The results demonstrate that, compared to the four typical deep regression models, the new model with dynamic feature-weighted fusion exhibits excellent performance in predicting outcomes across the entire firing boundary area, providing an intelligent solution for dynamic firing range estimation in weapon systems.