Real-time spectral target detection based on saliency and edge contour fusion

Snapshot spectral cameras have successfully achieved real-time spectral imaging. However, existing spectral target detection algorithms exhibit lengthy computation times, which do not satisfy the requirements for real-time processing. This limitation significantly hinders their application and wider adoption. To address this issue, a novel real-time spectral target detection method, which integrates saliency and edge contour detection, is proposed in this paper. Initially, dimensionality reduction of the spectral image data is performed based on spectral differences. Subsequently, a saliency detection algorithm is utilized, which calculates the Euclidean distance between pixel vectors in the LAB color space and the average pixel vector to identify salient regions. Following this, edge contour detection is conducted using the gradient magnitude of the image. Finally, a weighted fusion of the salient regions and edge contour detection results is implemented. Experimental results demonstrate that the proposed method achieves optimal detection performance across four distinct datasets, with AUC values surpassing 0.996 and exhibiting robust performance. Furthermore, the processing time is minimized, averaging only 0.0234 s. This advancement enables efficient real-time spectral target detection using snapshot multispectral cameras.