Fei Zhu

Transformers have shown significant success in hyperspectral unmixing (HU). However, challenges remain. Transformer-based unmixing networks, built on Vision Transformer (ViT) or Swin Transformer, struggle to effectively capture essential multi-scale and long-range spatial correlations. Additionally, these networks predominantly rely on the linear mixing model, lacking the flexibility to accommodate scenarios with significant nonlinear effects. To address these limitations, we propose a multi-scale Dilated Transformer-based unmixing network for nonlinear HU (DTU-Net). Its encoder integrates two branches: a spatial branch mainly employing Multi-Scale Dilated Attention (MSDA) to uniquely capture intricate multi-scale and long-range spatial correlations via adaptive receptive fields, and a spectral branch utilizing 3D-CNNs with channel attention. This design enables comprehensive extraction as well as integration of multi-level spatial and spectral features. The decoder is specifically designed to accommodate both linear and nonlinear mixing. It explicitly models the polynomial post-nonlinear mixing model (PPNMM) by learning nonlinear coefficients as pixel-wise features, which enhances interpretability by directly reflecting the pixel-level nonlinear mixing strength. Experiments on synthetic, ray tracing, and real datasets validate the effectiveness of the proposed DTU-Net, demonstrating its superior performance compared to both PPNMM-derived and advanced unmixing networks. The code is available at: https: //github.com/ChentongWang/DTU-Net.

DNA is a promising storage medium, offering advantages in high density, long durability, and low maintenance cost. However, information recovery in DNA storage systems is challenged by errors arising during synthesis, amplification, and sequencing phases. A key challenge in decoding is sequence reconstruction, which involves recovering the original reference sequence from a set of noisy copies. While recent research has explored deep learning-based methods for this task, the high cost of synthesis and sequencing results in a limited availability of training samples. To overcome this challenge, we propose TransDNA, a deep transfer learning network specifically designed for sequence reconstruction in DNA storage. It consists of an encoder, a domain-specific decoder, and a domain-invariant feature extractor, with alternating domain alignment and domain-specific reconstruction mechanisms. By transferring knowledge from a larger source dataset, TransDNA significantly enhances the reconstruction success rate on two target datasets from real DNA storage experiments, outperforming the base model without transfer learning and several comparative methods. Notably, TransDNA surpasses the SDG method in both reconstruction success rate and training efficiency. These results demonstrate the effectiveness of TransDNA as the first transfer learning approach applied to the DNA sequence reconstruction task. The source code is available at: https://github.com/qinyunnn/TransDNA.

Sparse unmixing, a critical task in hyperspectral image interpretation, aims to identify an optimal subset of end-members from a predefined library and estimate the fractional abundances for each pixel. However, in real-world scenarios, various types of noise significantly degrade the performance of conventional sparse unmixing methods that usually rely on ℓ₂-norm loss function. To address this issue, this letter proposes a robust sparse unmixing method based on the continuous mixed norm (CMN), which exhibits resilience to mixed noise, particularly non-Gaussian impulsive noise. By adopting CMN as the reconstruction loss function, we formulate both the standard sparse unmixing problem and its augmented version with total variation (TV) regularizer for spatially piecewise smoothness. The corresponding algorithms are derived using the alternating direction method of multipliers (ADMM). Experiments on both synthetic and real hyperspectral datasets validate the effectiveness and robustness of the proposed method in handling diverse and mixed noise conditions over comparing methods. The code is available at: https://github.com/JinchengGao/CMNSU.

Transformers have shown significant success in hyperspectral unmixing (HU). However, challenges remain. While multi-scale and long-range spatial correlations are essential in unmixing tasks, current Transformer-based unmixing networks, built on Vision Transformer (ViT) or Swin-Transformer, struggle to capture them effectively. Additionally, current Transformer-based unmixing networks rely on the linear mixing model, which lacks the flexibility to accommodate scenarios where nonlinear effects are significant. To address these limitations, we propose a multi-scale Dilated Transformer-based unmixing network for nonlinear HU (DTU-Net). The encoder employs two branches. The first one performs multi-scale spatial feature extraction using Multi-Scale Dilated Attention (MSDA) in the Dilated Transformer, which varies dilation rates across attention heads to capture long-range and multi-scale spatial correlations. The second one performs spectral feature extraction utilizing 3D-CNNs with channel attention. The outputs from both branches are then fused to integrate multi-scale spatial and spectral information, which is subsequently transformed to estimate the abundances. The decoder is designed to accommodate both linear and nonlinear mixing scenarios. Its interpretability is enhanced by explicitly modeling the relationships between endmembers, abundances, and nonlinear coefficients in accordance with the polynomial post-nonlinear mixing model (PPNMM). Experiments on synthetic and real datasets validate the effectiveness of the proposed DTU-Net compared to PPNMM-derived methods and several advanced unmixing networks.

DNA holds immense potential as an emerging data storage medium. However, the recovery of information in DNA storage systems faces challenges posed by various errors, including IDS errors, strand breaks, and rearrangements, inevitably introduced during synthesis, amplification, sequencing, and storage processes. Sequence reconstruction, crucial for decoding, involves inferring the DNA reference from a cluster of erroneous copies. While most methods assume equal contributions from all reads within a cluster as noisy copies of the same reference, they often overlook the existence of contaminated sequences caused by DNA breaks, rearrangements, or mis-clustering reads. To address this issue, we propose RobuSeqNet, a robust multi-read reconstruction neural network specifically designed to robustly reconstruct multiple reads, accommodating noisy clusters with strand breakage, rearrangements, and mis-clustered strands. Leveraging the attention mechanism and an elaborate network design, RobuSeqNet exhibits resilience to highly-noisy clusters and effectively deals with in-strand IDS errors. The effectiveness and robustness of the proposed method are validated on three representative next-generation sequencing datasets. Results demonstrate that RobuSeqNet maintains high sequence reconstruction success rates of 99.74%, 99.58%, and 96.44% across three datasets, even in the presence of noisy clusters containing up to 20% contaminated sequences, outperforming known sequence reconstruction models. Additionally, in scenarios without contaminated sequences, it exhibits comparable performance to existing models, achieving success rates of 99.88%, 99.82%, and 97.68% across the three datasets.

Unsupervised spectral unmixing consists of representing each observed pixel as a combination of several pure materials known as endmembers, along with their corresponding abundance fractions. Beyond the linear assumption, various nonlinear unmixing models have been proposed, with the associated optimization problems solved either by traditional optimization algorithms or deep learning techniques. Current deep learning-based nonlinear unmixing mainly focuses on additive, bilinear-based formulations. The multilinear mixing model (MLM) offers a unique perspective by interpreting the reflection process by discrete Markov chains, allowing it to account for interactions between endmembers up to infinite order. However, explicitly simulating the physics of MLM using neural networks has remained a challenging problem. In this paper, we propose a novel autoencoder-based network for unsupervised unmixing based on MLM. Leveraging an elaborate network design, this approach explicitly models the relationships among all model parameters: endmembers, abundances, and transition probability. The network operates in two modes: MLM-1DAE, which considers only pixel-wise spectral information, and MLM-3DAE, which explores spectral-spatial correlations within input patches. Experiments on both the synthetic and real datasets validate the effectiveness of the proposed method, demonstrating competitive performance compared to classic MLM-based solutions. The code is available at https://github.com/ting-Fang09/Hyperspectral-unmixing-MLM-AE.

Unsupervised spectral unmixing consists of representing each observed pixel as a combination of several pure materials called endmembers with their corresponding abundance fractions. Beyond the linear assumption, various nonlinear unmixing models have been proposed, with the associated optimization problems solved either by traditional optimization algorithms or deep learning techniques. Current deep learning-based nonlinear unmixing focuses on the models in additive, bilinear-based formulations. By interpreting the reflection process using the discrete Markov chain, the multilinear mixing model (MLM) successfully accounts for the up to infinite-order interactions between endmembers. However, to simulate the physics process of MLM by neural networks explicitly is a challenging problem that has not been approached by far. In this article, we propose a novel autoencoder-based network for unsupervised unmixing based on MLM. Benefitting from an elaborate network design, the relationships among all the model parameters {\em i.e.}, endmembers, abundances, and transition probability parameters are explicitly modeled. There are two modes: MLM-1DAE considers only pixel-wise spectral information, and MLM-3DAE exploits the spectral-spatial correlations within input patches. Experiments on both the synthetic and real datasets demonstrate the effectiveness of the proposed method as it achieves competitive performance to the classic solutions of MLM.