Deep learning-based hyperspectral unmixing (HSU) has followed a trajectory of increasing architectural complexity, from simple autoencoders to transformers and, most recently, Mamba-based models, each claiming to advance the state of the art. Contrary to this trend, we propose a lightweight CNN-based autoencoder, coined "Not New Autoencoder Unmixing" (nnAEU). It is derived from the backbone of UNMamba (2025) through a pruning study that removes the Mamba component entirely. nnAEU uses less than 10 times fewer parameters than UNMamba, and far fewer than other transformer- and GAN-based methods, while outperforming them in accuracy. We provide a complete, minimal, reproducible code, available at https://gitlab.telecom-paris.fr/ring/nnAEU.git.
Figures & tables
Fig. 1 : Comparison of UNMamba and proposed nnAEU unmixing networks. nnAEU is derived from UNMamba by pruning the UNMamba blocks and simplifying the endmember module. Y is the input HSI, A,E are the abundance and endmember estimations, Y is the reconstructed HSI and S is the trainable endmember sequence randomly initialised. Weight are trainable multiplicative factors and Avg. represents the averaging operation.
Samson
Jasper
Apex
Urban
Score ↓
SAD ↓
NMSE ↓
Time
Size
SAD ↓
NMSE ↓
Time
Size
SAD ↓
NMSE ↓
Time
Size
SAD ↓
NMSE ↓
Time
Size
UNMamba [ 3 ]
0.025 ± 0.01
0.014 ± 0
32
194k
0.095 ± 0.03
0.033 ± 0.04
40
242k
0.139 ± 0.06
0.116 ± 0.05
38
260k
0.165 ± 0.05
0.215 ± 0.13
202
3.8M
1.23
UNMamba-p
0.024 ± 0
0.016 ± 0
9
43k
0.065 ± 0.01
0.019 ± 0.01
12
62k
0.122 ± 0.02
0.096 ± 0.02
9
85k
0.148 ± 0.04
0.217 ± 0.09
83
69k
0.81
Table 1: Comparison of the performance of UNMamba and UNMamba-p, without the UNMamba blocks. Time is in seconds, Score is detailed in subsection 4.2 , and Size is the number of trainable parameters. Best results in bold , second best underlined
Fig. 2 : Comparison of all method’s estimated A on the Apex dataset, our proposed method is in bold .
Fig. 3 : Comparison of all method’s estimated E on the Apex dataset, our proposed method is in bold .
Samson
Jasper
Apex
Urban
Repr.
Rank
SAD ↓
NMSE ↓
Time
Size
SAD ↓
NMSE ↓
Time
Size
SAD ↓
NMSE ↓
Time
Size
SAD ↓
NMSE ↓
Time
Size
SiVM + FCLS [ 6 ]
0.056 ± 0
0.438 ± 0
0
0
0.261 ± 0
0.187 ± 0
0
0
0.244 ± 0
0.943 ± 0
0.1
0
0.436 ± 0
1.293 ± 0
0.1
0
✔
7
CNNAEU [ 13 ]
0.058 ± 0.08
0.112 ± 0.11
1.4
124k
0.253 ± 0.17
0.328 ± 0.16
1.4
182k
0.362 ± 0.24
0.263 ± 0.12
1.5
261k
0.238 ± 0.07
0.712 ± 0.24
5.4
188k
✗
5
UnDIP [ 15 ]
0.056 ± 0
0.453 ± 0
57
1.6M
0.261 ± 0
0.191 ± 0
72
1.7M
0.244 ± 0
1.003 ± 0
108
1.9M
0.436 ± 0
1.242 ± 0.01
269
1.6M
✗
8
DeepTrans [ 5 ]
0.198 ± 0.01
0.346 ± 0
3.2
5M
0.282 ± 0.01
0.255 ± 0.01
3.2
7.5M
0.289 ± 0
0.445 ± 0.01
3
8M
0.369 ± 0.03
0.786 ± 0.11
9
34M
✗
10
SSST-GAN [ 20 ]
0.136 ± 0.01
0.078 ± 0.03
173
175M
0.304 ± 0.06
0.259 ± 0.04
172
175M
0.380 ± 0.09
0.808 ± 0.03
179
175M
0.334 ± 0.09
0.617 ± 0.17
175
175M
✔
9
Table 2: Comparison of all HSU methods on the 4 datasets, averaged over 10 runs. We display results as Mean ± std , Time (in seconds), Rank (as described in 4.2 ), Size (number of trainable parameters), and Repr. indicates whether the method’s announced performance could be replicated or not.
Hyperspectral unmixing (HU) is crucial for analyzing hyperspectral imagery, yet achieving accurate unmixing remains challenging. While traditional methods struggle to effectively model complex spectral-spatial features, deep learning approaches often lack physical interpretability. Unrolling-based methods, despite offering network interpretability, inadequately exploit spectral-spatial information and incur high memory costs and numerical precision issues during backpropagation. To address these limitations, we propose DEQ-Unmix, which reformulates abundance estimation as a deep equilibrium model, enabling efficient constant-memory training via implicit differentiation. It replaces the gradient operator of the data reconstruction term with a trainable convolutional network to capture spectral-spatial information. By leveraging implicit differentiation, DEQ-Unmix enables efficient and constant-memory backpropagation. Experiments on synthetic and two real-world datasets demonstrate that DEQ-Unmix achieves superior unmixing performance while maintaining constant memory cost.
Chentong Wang, Jincheng Gao, Fei Zhu +1
Center for Applied Mathematics, KL-AAGDM, Tianjin University, Tianjin 300072, China · School of Artificial Intelligence, Northwestern Polytechnical University, Xi’an 710072, China · National Engineering Laboratory for Integrated Aero-Space-Ground-Ocean Big Data Application Technology
Several foundation models dedicated to hyperspectral images have recently been made available. These models are trained on large unlabeled datasets and exhibit strong performance on many hyperspectral imaging tasks, such as classification or denoising. Nonetheless, their performance for hyperspectral unmixing -- the task of separating mixed spectra of overlapping materials in a hyperspectral image -- remain understudied. This might partly be due to the fact that most of them rely on vision transformer backbones, including patchification, leading to a feature resolution problem. While hyperspectral unmixing already arises from the low resolution of hyperspectral images, this patchification step potentially makes the problem even more ill-posed. Therefore, in this work, we aim to answer two questions: 1) how do foundation models perform in hyperspectral unmixing?; 2) how to tackle the feature-level loss of resolution? To answer the first question, we benchmark foundation models for unmixing, showing that they can reach state-of-the-art performance on four hyperspectral unmixing datasets. To answer the second question, we compare several feature upsampling approaches and empirically show that using a simple one can lead to high performance results. The code is available at https://gitlab.telecom-paris.fr/ring/hfm-hsu.git.
Edgard Dabier, Christophe Kervazo, Pietro Gori +1
LTCI, T´el´ecom Paris, Institut Polytechnique de Paris, Palaiseau, France
Due to the large footprint of pixels in remote sensing imagery, hyperspectral unmixing (HU) has become an important and necessary procedure in hyperspectral image analysis. Traditional HU methods rely on a prior spectral mixing model, especially for nonlinear mixtures, which has largely limited the performance and generalization capacity of the unmixing approach. In this paper, we address the challenging problem of hyperspectral nonlinear unmixing (HNU) without explicit knowledge of the mixing model. Inspired by the principle of generative models, where images of the same distribution can be generated as that of the training images without knowing the exact probability distribution function of the image, we develop an invertible mixing-unmixing process via a bi-directional GAN framework, constrained by both the cycle consistency and the linkage between linear and nonlinear mixtures. The combination of cycle consistency and linear linkage provides powerful constraints without requiring an explicit mixing model. We refer to the proposed approach as the linearly-constrained CycleGAN unmixing net, or LCGU net. Experimental results indicate that the proposed LCGU net exhibits stable and competitive performance across different datasets compared with other state-of-the-art model-based HNU methods.
Maofeng Tang, Hairong Qi
Department of Electrical Engineering and Computer Science, University of Tennessee, Knoxville