Estimating the distribution of wood stove appliances is essential for fine-scale air quality management but remains limited by the coarse spatial resolution of survey-based inventories such as the U.S. National Emissions Inventory (NEI). To address this limitation, we proposed a novel computer vision framework that leverages visible chimney features as proxies for wood stove usage. Using data from drones, vehicle-based videos across urban and rural areas, we trained object detection models (YOLOv11) and integrated them with Vision Language Models (VLMs) for semantic verification. This two-stage pipeline substantially improves detection accuracy over YOLO alone. In the urban Fall Creek neighborhood, the YOLO+VLMs approach achieved an F1-score of 61.4%, compared to 29.3% using YOLO alone. In rural video data, the framework reached 74.5%, up from 17.5%. These results demonstrate the viability of context-informed, multimodal models in identifying distributed emission sources. The proposed method is scalable and adaptable, offering a promising tool for real-time residential wood combustion mapping and environmental policy development.