Electrical impedance tomography (EIT) is one of the typical ill-posed inverse problems, where serious ill-posedness and the linear approximation of the forward operator lead to obvious distortions and artifacts in the...
详细信息
Electrical impedance tomography (EIT) is one of the typical ill-posed inverse problems, where serious ill-posedness and the linear approximation of the forward operator lead to obvious distortions and artifacts in the degraded reconstructions, further limiting its practical application. The learning-based strategies with image enhancement have been introduced into EIT reconstruction and also achieved improvements. Nevertheless, this idea ignores the priori knowledge of physical information, while not fully exploiting data consistency, resulting in poor generalization and interpretability. In this work, a reweighted Split Bregman (SB) iterative algorithm is proposed regularized by total variation firstly, referred to as RwTVSB. Moreover, the RwTVSB iteration is unrolled into a neural network-based learning framework, dubbed as RwTVSB-Net. The reweighted matrix is introduced to the SB iteration, which could overcome the loss of information of the forward operator due to the linear approximation and also enhance the constraints of the physical priori. Specifically, (1) a network based on residual connection and SE-attention is designed to update the reweighted matrix. (2) Further, a U-shaped architecture with deformable large kernel convolution, dilated convolution, and cross-attention is embedded into this unrolling framework to learn the soft threshold operator. This not only maintains consistency with the RwTVSB iterative algorithm but also uses multi-scale features to fusion information at multiple levels. Both simulated and real-world measured data are employed to validate the effectiveness and advantages of the proposed RwTVSB-Net. The visual reconstructions and quantitative metrics show that RwTVSB-Net outperforms other state-of-the-art methods. In addition, the robustness of the method is tested and validated on multiple imaging tasks.
At present, an emerging technique called the algorithmunrolling approach has attracted wide attention, because it is capable of developing efficient and interpretable layers to eliminate the black-box nature of deep ...
详细信息
At present, an emerging technique called the algorithmunrolling approach has attracted wide attention, because it is capable of developing efficient and interpretable layers to eliminate the black-box nature of deep learning (DL). In this article, inspired by the sparse unmixing model, we propose a model-driven DL approach, namely, an implicit variable iterative unrolling network (IVIU-Net). First of all, the unmixing performance and adaptive ability of the model are enhanced by introducing learnable parameters into the sparse unmixing algorithm. Then, a specific spatial convolution module is integrated into the network to promote the smoothness of the latent abundance map. Finally, a comprehensive loss function with three terms such as average spectral angle distance, hyperspectral images reconstruction error, and spectral information divergence, is presented to train the IVIU-Net in an unsupervised way. Compared to the unmixing results of most existing data-driven DL algorithms, our network has significant advantages in two folds: it is able to achieve better stability instead of relying heavily on the endmember initialization results and it has better interpretability and robustness in the unmixing procedure. Experimental results on synthetic and real data show that the proposed network outperforms the state-of-the-art in terms of better convergence, faster unmixing speed as well as better accuracy.
Multispectral remote-sensing images often have band-dependent image resolution due to cost and technical limitations. To address this, we developed a method that sharpens low-resolution (LR) images using high-resoluti...
详细信息
Multispectral remote-sensing images often have band-dependent image resolution due to cost and technical limitations. To address this, we developed a method that sharpens low-resolution (LR) images using high-resolution (HR) images. In this letter, we propose a novel unsupervised deep-learning (DL) approach that involves unrolling an iterative algorithm into a deep neural network and training it using a loss function based on Stein's risk unbiased estimate (SURE) to sharpen the LR bands (20 and 60 m) of Sentinel-2 (S2) to their highest resolution (10 m). This approach views traditional optimization model-based methods through a DL framework, improving interpretability and clarifying connections between the two approaches. Results from both simulated and real S2 datasets demonstrate that the proposed method outperforms competitive methods and produces high-quality sharpened images for the 20- and 60-m bands. The codes are available at https://***/hvn2/S2-unrolling.
Due to its all-day and all-weather capability, synthetic aperture radar (SAR) plays an important role in many remote sensing and monitoring applications. However, conventional SAR image reconstruction methods generall...
详细信息
Due to its all-day and all-weather capability, synthetic aperture radar (SAR) plays an important role in many remote sensing and monitoring applications. However, conventional SAR image reconstruction methods generally perform undifferentiated imaging, complicating target detection. To address this challenge, we propose a SAR image reconstruction method based on self-attention deep prior learning for differentiated image reconstruction and target detection. The proposed method can separate the target from the clutter using their feature priors during image reconstruction, thus helping improve target detection performance. Specifically, a deep prior learning operation based on a self-attention convolutional neural network (SACNN) is proposed. SACNN can help enhance target and suppress clutter by learning both the local and global features. Finally, the proposed method is implemented through an unrolled deep network with a loss function designed to make the reconstructed image beneficial for target detection. Simulation experiments have been conducted to verify the efficacy of the proposed method.
暂无评论