Research on dangerous goods sign recognition based on multimodal feature fusion

Accurate recognition of dangerous goods signs is crucial for safety in international cargo transport and emergency management. This paper proposes an intelligent method using multi-feature fusion to overcome traditional limits in complex backgrounds, lighting, and occlusion. The utilization of the HSV color space, threshold segmentation, and morphological optimization is instrumental in ensuring the precise positioning of the designated sign area. A multi-dimensional descriptor is constructed, integrating color histogram, edge density, and grayscale symbology matrix texture features. A weighted similarity-matching strategy is employed to facilitate fast retrieval and classification decisions from the template library. Experiments on a dataset of 1200 test images show that the system meets the accuracy requirements for 9 classes and 20 divisions of hazard sign recognition. This can be effectively applied to safety supervision, fire emergency scenarios, and other similar contexts, providing support for the development of intelligent ports.