Typical battlefield infrared background detection method based on multi band fusion

Intelligent battlefield environment recognition is crucial for active camouflage technology. Enhancing detection capabilities for various environments is essential for target survival. Traditional systems, relying on single visible light or infrared bands, face challenges like low detection performance and limited information use due to lighting conditions, making them inadequate for all-weather detection. This study presents a multi-modal feature fusion network model using a typical background database. It employs a coordinated attention mechanism for spatial information and optimizes dense and dual-path networks to improve the fusion of optical and infrared images. The model achieves 97.57% accuracy, 4.16% higher than the best single-modal results. The attention mechanism boosts accuracy by 2.68%. Thus, the model effectively integrates optical and infrared data, showing strong performance in classifying and detecting typical battlefield backgrounds. Article Highlights A multimodal feature fusion model is proposed based on optical and infrared images to achieve accurate detection of different background types. This article starts from heterogeneous data and constructs a typical background database by building a multimodal data acquisition platform. Optical and infrared data are aligned through image cropping and correction. By introducing the CA attention mechanism to optimize the network model, feature extraction was performed on two types of heterogeneous data, improving the feature extraction capability. The detection of typical background types through multimodal fusion technology provides an effective approach for the research of detecting other different surface types.