The Application of Parameter-Efficient Fine-Tuning with Frequency Filter Adapter in Urban Governance

With the continuous advancement of urbanization and the ongoing improvements in multimodal large model technologies, it is imperative to develop an automated, multimodal industry-specific large model for handling a substantial volume of urban management cases, thereby enhancing the level of intelligent urban governance. After fine-tuning, multimodal large models can replace human labor to a certain extent in performing diverse tasks, demonstrating effective results in addressing urban governance issues. However, multimodal large models trained with existing parameter-efficient fine-tuning techniques, such as LoRA, still fall short of expectations when it comes to specialized performance in complex urban management scenarios. To bridge this gap, we propose a Frequency Filter Adapter (FFA) fine-tuning technique. This module extracts features from the frequency domain of urban management scene images, providing richer visual information to the multimodal large model at the same image resolution by utilizing the frequency domain characteristics. Experimental results indicate that the use of the Frequency Filter Adapter (FFA) effectively enhances the multi-modal large model's understanding of image details within urban management scenarios, leading to more accurate judgments and intelligent decision-making, thus significantly elevating the level of intelligence in urban management. Finally, we offer forward-looking thoughts on the application prospects of multimodal large models in the field of urban management.