BS-Mamba for Black-Soil Area Detection On the Qinghai-Tibetan Plateau
Xuan Ma, Zewen Lv, Chengcai Ma, Tao Zhang, Yuelan Xin, Kun Zhan1. School of Information Science and Engineering, Lanzhou University2. State Key Laboratory of HIGAE, College of Ecology, Lanzhou University3. School of Physics and Electronic Information Engineering, Qinghai Normal University4. Key Labo · 2025
Read BS-Mamba for Black-Soil Area Detection On the Qinghai-Tibetan Plateau by Xuan Ma, Zewen Lv, Chengcai Ma, Tao Zhang, Yuelan Xin, Kun Zhan1. School of Information Science and Engineering, Lanzhou University2. State Key Laboratory of HIGAE, College of Ecology, Lanzhou University3. School of Physics and Electronic Information Engineering, Qinghai Normal University4. Key Labo (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-12495.txt.
Introduction
Xuan Ma and co-authors build a data set and a new model for spotting badly degraded grassland on the Qinghai-Tibetan Plateau, in a paper from 2025. Such land, called black soil area, loses plant cover and soil quality through overgrazing, climate change, and rodent activity. Mapping it well is a first step toward repair.
The team gathered drone images of the plateau and labeled them with expert guidance. Their model, named BS-Mamba, is built to read those images and mark the degraded patches. On two separate test sets it beat the best current models in accuracy. The authors present the data set and model as tools that land managers can use to judge how much land needs restoration work.
Catalog record
| CREATOR | Xuan Ma, Zewen Lv, Chengcai Ma, Tao Zhang, Yuelan Xin, Kun Zhan1. School of Information Science and Engineering, Lanzhou University2. State Key Laboratory of HIGAE, College of Ecology, Lanzhou University3. School of Physics and Electronic Information Engineering, Qinghai Normal University4. Key Labo |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 43 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-12495
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
BS-Mamba for Black-Soil Area Detection On the Qinghai-Tibetan Plateau Xuan Ma, Zewen Lv, Chengcai Ma, Tao Zhang, Yuelan Xin, Kun Zhan1. School of Information Science and Engineering, Lanzhou University2. State Key Laboratory of HIGAE, College of Ecology, Lanzhou University3. School of Physics and Electronic Information Engineering, Qinghai Normal University4. Key Laboratory of AI and Information Processing, Hechi University https://github.com/kunzhan/BS-Mamba Abstract Extremely degraded grassland on the Qinghai-Tibetan Plateau (QTP) presents a significant environmental challenge due to overgrazing, climate change, and rodent activity, which degrade vegetation cover and soil quality. These extremely degraded grassland on QTP, commonly referred to as black-soil area, require accurate assessment to guide effective restoration efforts. In this paper, we present a newly created QTP black-soil dataset, annotated under expert guidance. We introduce a novel neural network model, BS-Mamba, specifically designed for the black-soil area detection using UAV remote sensing imagery. The BS-Mamba model demonstrates higher accuracy in identifying black-soil area across two independent test datasets than the state-of-the-art models. This research contributes to grassland restoration by providing an efficient method for assessing the extent of black-soil area on the QTP. 1 Introduction The Qinghai-Tibetan Plateau (QTP) is an ecological region that significantly influences climate regulation and water security in Asia, while also global carbon cycling and serves as a sanctuary for numerous unique species [ 1 ] . However, QTP has faced increasing concern due to extremely grassland degradation. Restoring these degraded grassland, such as the black-soil-type areas on the QTP, is a significant challenge, often requiring over two centuries to restore ecosystem health [ 3 ] . Black-soil patches, characterized by the absence of mattic epipedon [ 22 ] , are recognized as a unique ecological phenomenon that primarily occurs within the grasslands of QTP [ 23 ] . The distribution of black-soil area on the QTP is expanding, primarily concentrated in the cold and extreme environments at altitudes ranging between 3,000 and 5,000 meters [ 23 ] . Black-soil area detection inherently involves identifying the extent of mattic epipedon, as the two conditions are mutually exclusive and collectively exhaustive. Given this binary nature of the relationship between black-soil area and mattic epipedon, we have developed a QTP black-soil dataset using the unmanned aerial vehicle (UAV) imagery, carefully annotating black-soil patches and mattic epipedon under the guidance of ecological experts. Exploiting this dataset, we train a model to infer black-soil area, facilitating the evaluation of restoration strategies and monitoring grassland quality on the QTP. To enhance detection capabilities, we aim to design a binary segmentation neural network specifically for black-soil area detection. U-Net [ 20 , 11 , 34 ] is widely recognized for its outstanding performance in binary segmentation tasks. While vanilla U-Net [ 20 ] excels at extracting local pixel-wise features through convolutional operations, it has limitations in capturing long-range dependencies between correlated pixels within the same class. In contrast, Mamba [ 6 ] , which utilizes patch-based sequence processing, are particularly effective at capturing long-range dependencies between different patches. Recent advancements in vision Mamba [ 33 , 31 , 35 ] have shown superior performance over both convolutional neural networks and vision transformers [ 4 ] in many computer vision tasks. Mamba, known for its efficiency in handling long sequences and capturing global contextual information as a state-space model (SSM) [ 6 ] , has emerged as a highly effective alternative to transformers. Thus, vision Mamba models have achieved state-of-the-art performance in various computer vision tasks. Building on recent innovations of vision Mamba [ 33 , 31 , 35 ] and hybrid structures, we introduce BS-Mamba, a novel black-soil area detection network that leverages the strengths of the Mamba [ 6 ] and incorporates skip connections from multi-branch encoder features to the decoder, enhancing detection performance overall. In this paper, we propose a BS-Mamba model, which integrates a multi-branch parallel feature fusion module into Mamba. This model combines low-level and high-level features by injecting semantic information into the low-level features and refining the high-level ones with detailed spatial information. The multi-branch layer-wise parallel structure compensates for the loss of feature resolution typically seen in Mamba. Inspired by the multi-branch layer-wise parallel design, the multi-scale features extracted by Mamba are upsampled and combined with the different-resolution features from the encoder, allowing for precise exploring long-range dependencies. Figure 1 : Comparison of our BS-Mamba model and a variant of Mamba, Mamba-Unet, in detecting black-soil areas. In both gravel-type and poisonous-plant-type regions, our BS-Mamba model consistently outperforms Mamba-Unet. Detection challenges include omissions, inaccuracies, and unclear boundaries. These issues arise mainly due to the similar color tones between black-soil patches and mattic epipedon (first row) and the high diversity of invasive weeds (second row). The results demonstrate that our BS-Mamba model effectively mitigates these challenges. Figure 1 illustrates the black-soil area detection results of our BS-Mamba model compared to a Mamba variant, Mamba-UNet [ 29 ] . Detection with Mamba variants faces challenges due to several factors: (1) the varying stages of degradation introduce significant diversity among black-soil types, complicating accurate identification; (2) the natural boundaries between grasslands and black-soil patches are often unclear. For instance, gravel-type black-soil patches [ 36 ] seamlessly blend with surrounding grassy areas, as shown in the first row of Figure 1 , leading to misclassifications and omissions. Additionally, dense weed cover, the presence of poisonous plants [ 21 ] , or the smooth surfaces of weed-covered areas further impair detection accuracy, as depicted in the second row of Figure 1 . In contrast, the third column of Figure 1 demonstrates that our BS-Mamba model effectively addresses these challenges, achieving significant improvements in accuracy and boundary clarity. BS-Mamba retains the strengths of Mamba while enhancing the segmentation of the black-soil patches and mattic epipedon. Empirical results show that our hybrid architecture, which integrates convolutional blocks with Mamba blocks, outperforms methods based solely on standalone convolutional blocks or Mamba blocks. This hybrid approach effectively combines the high-resolution spatial features from convolutional blocks with the global context captured by Mamba. Unlike Mamba-UNet, which only uses Mamba blocks within a U-shaped network architecture, our BS-Mamba model fuses both convolutional and Mamba features within a two-branch encoder framework. This design enables more accurate and efficient detection of black-soil areas. We conduct extensive experiments on the proposed QTP-BS dataset, showing the outstanding performance of the BS-Mamba model in precise black-soil patches and mattic epipedon segmentation tasks. To further validate the model’s generalization ability, we also perform comprehensive testing on a secondary test set. The results conclusively demonstrate the strong adaptability and stability of the BS-Mamba model under varying environmental conditions. The main contributions are summarized as follows: (1) We have assembled a large-scale dataset of black-soil grasslands on the QTP, named the QTP-BS dataset. To the best of our knowledge, this is the first dataset created specifically for the challenging task of black-soil area detection. The QTP-BS dataset includes 15,000 images of classic black-soil areas in the QTP grasslands, along with their corresponding ground-truth labels. (2) We proposed a BS-Mamba model, which innovatively combines Mamba and convolution, leveraging the high-resolution spatial details of convolution features and the long-range dependencies encoded by Mamba. Experiments demonstrate that BS-Mamba performs exceptionally well in black-soil patches and mattic epipedon detection. (3)