Fusion visual saliency with dual-spectral saliency of anomaly detection for hyperspectral herbal oral liquid

Ating Yin,Yaping Li,Kai Zeng,Yaonan Wang,Jianxu Mao,Hui Zhang

DOI: https://doi.org/10.1117/12.2664199

2023-03-20

Abstract:Chinese herbal oral liquid can leach a variety of effective ingredients from herbs and has become a major drug for clinical application. However, it is easy to produce or introduce foreign matters that are very faint in the automatic filling production process. To solve the challenge of low accuracy of faint foreign matter detection, in this paper, we proposed a salient-based anomaly detection method which is fuses visual saliency with dual-spectral saliency (VDS) for the hyperspectral herbal oral liquid. Specifically, we first select the most discriminative bands via the band selection method to generate the pseudo-color map. Subsequently, the histogram-based contrast method is introduced to select the saliency feature map with the largest variance of color features, while fusing the multi-scale gradient features to obtain the preliminary vision-based anomaly detection map. After that, the spectral angles and spectral Euclidean distances are calculated separately based on the oral liquid hyperspectral images to fused into dual-spectral saliency maps. Finally, the dual-spectral saliency map is employed to suppress the background information of the preliminary anomaly detection map. The experimental results show that our proposed method outperforms the state-of-the-art anomaly detection methods, which accurately and quickly achieve the detection of faint foreign matter in the hyperspectral herbal oral liquid. It will accelerate the process of automated filling production lines for oral liquid in the pharmaceutical industry.

Medicine,Engineering,Chemistry