Salient object detection on hyperspectral images has made some progress in recent years, benefiting from the development of wireless network and hyperspectral imaging technology. However, most object detection methods on hyperspectral images focus more on the spectrum and do not fully mine the spatial information, especially high-level spatial–spectral information. In this paper, we propose a salient object detection model on hyperspectral images in wireless network by applying saliency optimization to convolutional neural network (CNN) features. In the model, we firstly use CNN with two channels to extract spatial and spectral features of the same dimension respectively and conduct feature fusion at the end. Then, we generate the final saliency maps by optimizing the saliency values of the foreground and background cues, computing from the CNN features. The experimental results confirm that the proposed method is effective and has better performance on hyperspectral images.

得益于无线网络和高光谱成像技术的发展，近年来高光谱图像上的显着目标检测取得了一些进展。但是，大多数高光谱图像上的目标检测方法更多地集中在光谱上，并未充分挖掘空间信息，尤其是高级空间光谱信息。在本文中，我们通过将显着性优化应用于卷积神经网络（CNN）特征，提出了无线网络中高光谱图像的显着目标检测模型。在模型中，我们首先使用带有两个通道的CNN分别提取相同维数的空间和光谱特征，最后进行特征融合。然后，我们通过优化前景和背景提示的显着性值来生成最终显着性图，从CNN功能进行计算。实验结果证明，该方法是有效的，并且在高光谱图像上具有更好的性能。