A Path Planning Model for Urban Firefighting Integrating Visual Geometry Group and Deep Q-network

Existing works that integrates traditional convolutional neural networks (CNNs) with Deep Q-networks (DQNs) to address the challenges of image recognition and path planning for agents navigating within complex environments. However, traditional CNN have limitations in capturing high-level semantic features of environmental images. To tackle this challenge, this paper introduces a model that integrates the Visual Geometry Group (VGG) network with DQNs for path planning, specifically tailored for the urban firefighting domain. This paper first presents an overall framework for integrated VGG and DQN method for the urban firefighting. Subsequently, two fire feature extraction models with the VGG and the ResNet are developed to effectively capture critical visual information. Building on this foundation, the path planning decision-making model utilizing DQN is further enhanced to optimize navigation in complex urban firefighting scenarios. Finally, comprehensive simulation experiments are conducted to validate the feasibility and ef- fectiveness of the proposed model, demonstrating its potential for practical application in urban firefighting operations. We improved the traditional DQN algorithm using VGG and ResNet respectively, and conducted a comparative experiment. In com- parison with the ResNet+DQN path planning method, the model proposed in this paper achieves notable improvements in two critical metrics: the reward value and the number of iterative steps. Specifically, the reward value is enhanced by $28.8 \%$, while the number of iterative steps is reduced by $5 \%$. These enhancements are expected to significantly boost path planning efficiency in urban firefighting operations and provide a robust technical reference for decision-making in this domain.