Remote Sensing Weak Target Joint Detection Through Multimodal Feature Fusion

In the realm of remote sensing weak target reconnaissance missions, existing target detection algorithms often struggle to accurately extract spatial features of small and faint targets due to limitations posed by low spatial resolution and high signal noise in hyperspectral images captured by space-borne and air-borne sensors. Additionally, hyperspectral data pertaining to sensitive targets is often scarce and derived from disparate sources, introducing heterogeneity that traditional hyperspectral detection algorithms struggle to handle. To overcome these problems, this paper proposes a multimodal fusion method that combines hyperspectral images and high-resolution optical images. This method first maps heterogeneous hyperspectral data to common speatral bands to achieve unified representation and uses multi-instance learning technology to learn the spectral signature of the target to make up for the lack of pixel-level annotation. Meanwhile, a high-resolution optical dataset containing targets was constructed, and the YOLO target detection model was trained to capture the spatial features of the target. Finally, the decision level integrated hyperspectral and optical modes, and the results validated on test data show that this fusion strategy significantly improves the detection effect and accuracy of weak targets in remote sensing. Experimental results show that the method proposed in this paper has significant advantages in the field of remote sensing weak target detection.