Liang Hong

Prof. Liang Hong

School of Information Science and Technology, Yunnan Normal University, Kunming 650500, China

Research Focus & Metrics

Academic Career Length

8Years (2019-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
24 February 2020 in Sensors
Verified
Journal Article
Open access
PDF file
Haifeng LiHaifeng Li
Haifeng Li
HJHao Jiang
Hao Jiang
XGXin Gu
Xin Gu
JPJian Peng
Jian Peng
WLWenbo Li
Wenbo Li
Chao TaoChao Tao
Chao Tao
Remote sensing image scene classification has a high application value in the agricultural, military, as well as other fields. A large amount of remote sensing data is obtained every day. After learning the new batch data, scene classification algorithms based on deep learning face the problem of catastrophic forgetting, that is, they cannot maintain the performance of the old batch data. Therefore, it has become more and more important to ensure that the scene classification model has the ability of continual learning, that is, to learn new batch data without forgetting the performance of the old batch data. However, the existing remote sensing image scene classification datasets all use static benchmarks and lack the standard to divide the datasets into a number of sequential learning training batches, which largely limits the development of continual learning in remote sensing image scene classification. First, this study gives the criteria for training batches that have been partitioned into three continual learning scenarios, and proposes a large-scale remote sensing image scene classification database called the Continual Learning Benchmark for Remote Sensing (CLRS). The goal of CLRS is to help develop state-of-the-art continual learning algorithms in the field of remote sensing image scene classification. In addition, in this paper, a new method of constructing a large-scale remote sensing image classification database based on the target detection pretrained model is proposed, which can effectively reduce manual annotations. Finally, several mainstream continual learning methods are tested and analyzed under three continual learning scenarios, and the results can be used as a baseline for future work.

29 May 2019 in Sustainability
Verified
Journal Article
Open access
PDF file
RYRongfeng Yang
Rongfeng Yang
Yi LuoYi Luo
Yi Luo
KYKun Yang
Kun Yang
LHLiang Hong
Liang Hong
XZXiaolu Zhou
Xiaolu Zhou
Myanmar, abundant in natural resources, is one of the countries with high forest cover in Southeast Asia. Along with its rapid socio-economic development, however, the construction of large-scale infrastructure, expansion of agricultural land, and an increasing demand for timber products have posed serious threats to the forests and significantly affected regional sustainable development. However, the geographical environment in Myanmar is complex, resulting in the lack of long-term sequence of land cover data products. Based on 30 years’ Landsat satellite remote sensing imagery data and the land cover data extracted by a mixed classification method, this paper examined the spatial and temporal evolution characteristics of forest cover in Myanmar and investigated driving factors of the spatio-temporal evolution. Results show that the forest cover has decreased by 110,621 km2 in the past 30 years with the annual deforestation rate of 0.87%. Cropland expansion is the main reason for the deforestation throughout the study period. The study can provide basic information of the forest cover data to the Myanmar government for ecological environment protection. At the same time, it can provide important support to the “Belt and Road” initiative to invest in the region’s economy.

17 February 2021 in Remote Sensing
Verified
Journal Article
Open access
PDF file
JPJian Peng
Jian Peng
XMXiaoming Mei
Xiaoming Mei
WLWenbo Li
Wenbo Li
LHLiang Hong
Liang Hong
BSBingyu Sun
Bingyu Sun
Haifeng LiHaifeng Li
Haifeng Li
Scene understanding of remote sensing images is of great significance in various applications. Its fundamental problem is how to construct representative features. Various convolutional neural network architectures have been proposed for automatically learning features from images. However, is the current way of configuring the same architecture to learn all the data while ignoring the differences between images the right one? It seems to be contrary to our intuition: it is clear that some images are easier to recognize, and some are harder to recognize. This problem is the gap between the characteristics of the images and the learning features corresponding to specific network structures. Unfortunately, the literature so far lacks an analysis of the two. In this paper, we explore this problem from three aspects: we first build a visual-based evaluation pipeline of scene complexity to characterize the intrinsic differences between images; then, we analyze the relationship between semantic concepts and feature representations, i.e., the scalability and hierarchy of features which the essential elements in CNNs of different architectures, for remote sensing scenes of different complexity; thirdly, we introduce CAM, a visualization method that explains feature learning within neural networks, to analyze the relationship between scenes with different complexity and semantic feature representations. The experimental results show that a complex scene would need deeper and multi-scale features, whereas a simpler scene would need lower and single-scale features. Besides, the complex scene concept is more dependent on the joint semantic representation of multiple objects. Furthermore, we propose the framework of scene complexity prediction for an image and utilize it to design a depth and scale-adaptive model. It achieves higher performance but with fewer parameters than the original model, demonstrating the potential significance of scene complexity.

28 June 2022 in Geocarto International
Verified
Journal Article
Closed access
HLHe Li
He Li
LHLiang Hong
Liang Hong
As one of the world’s biodiversity hotspots, Mekong Basin has experienced a dramatic change in land-use and land-cover (LULC) over the past three decades, which could have a profound effect on regional biodiversity patterns. However, LULC dynamic and their driving forces in the Mekong Basin are still not well understood due to the complex geographical environment and the lack of efficient approaches to consistently produce accurate and timely LULC maps. In this study, we divided the Mekong Basin into three geographical units based on topographic and landscape heterogeneity. Then a rule-based method was employed to map LULC for each geographical unit from Landsat images. Finally, LULC dynamics and associated driving forces in the Mekong Basin from 1988 to 2017 were analysed. The results showed that: (1) the rule-based classification framework could accurately produce time series of LULC maps with overall accuracies over 86% for all single-year classifications. (2) Forest (49.92%) and cultivated land (34.89%) were the dominant LULC types in the Mekong Basin during the study period. Between 1988 and 2017, about 20.98% of the study area experienced LULC changes, in which the artificial surface increased by 3980.31 km2, mainly at the expense of forest and cultivated land. (3) Human disturbances were identified as the major driving forces of LULC changes. This study present essential source of information for policymakers to formulate sustainable natural resources management policies in the study site. In addition, this rule-based classification provides an insight into improving the accuracy of LULC mapping in complex geographical environment.

25 March 2024 in Water
Verified
Journal Article
Open access
PDF file
XYXiaozhou Yang
Xiaozhou Yang
LHLiang Hong
Liang Hong
Surface water is a critical natural resource, but its mapping accuracy is vulnerable to cloud cover, snow, shadows, and diverse roofing materials. Recognizing the limitations of a single threshold segmentation method that fails to achieve high-precision extraction of surface water in complex terrain areas, this study introduces a multiple threshold water detection rule (MTWDR) method to improve water extraction results. This method uses the multi-band reflectance characteristics of ground features to construct a water index and combines brightness features with the Otsu algorithm to eliminate interference from highly reflective ground features like ice, snow, bright material buildings, and clouds. The Yunan–Guizhou Plateau was selected as the study area due to its complex terrain and multiple types of surface water, and experiments were conducted using Sentinel-2 data on the Google Earth Engine (GEE). The results demonstrate that: (1) The proposed method achieves an overall accuracy of 94.08% and a kappa coefficient of 0.8831 in mountainous areas. In urban areas, the overall accuracy reaches 95.15%, accompanied by a kappa coefficient of 0.8945. (2) Compared to five widely used water indexes and rules, the MTWDR method improves accuracy by more than 3%. (3) It effectively overcomes interference from highly reflective ground features while maintaining the integrity and accuracy of water boundary extraction. In conclusion, the proposed method enhances extraction accuracy across different types of surface water within complex terrain areas, and can provide significant theoretical implications and practical value for researching and applying surface water resources.

6 March 2025 in Land
Verified
Journal Article
Open access
PDF file
ZYZanxian Yang
Zanxian Yang
FYFei Yang
Fei Yang
YXYuanjing Xiang
Yuanjing Xiang
HYHaiyi Yang
Haiyi Yang
CDChunnuan Deng
Chunnuan Deng
Zhongchang SunZhongchang Sun
Zhongchang Sun
The United Nations Sustainable Development Goal (SDG) 11.1 emphasizes improving well-being, ensuring housing security, and promoting social equity. Informal settlements, one of the most vulnerable groups, require significant attention due to their dynamic changes and habitat quality. These areas limit the ability to comprehensively capture spatial heterogeneity and dynamic shifts in regional sustainable development. This study proposes an integrated approach using multi-source remote sensing data to extract the spatial distribution of informal settlements in Mumbai and assess their habitat environment quality. Specifically, seasonal spectral indices and texture features were constructed using Sentinel and SAR data, combined with the mean decrease impurity (MDI) indicator and hierarchical clustering to optimize feature selection, ultimately using a random forest (RF) model to extract the spatial distribution of informal settlements in Mumbai. Additionally, an innovative habitat environment index was developed through a Gaussian fuzzy evaluation model based on entropy weighting, providing a more robust assessment of habitat quality for informal settlements. The study demonstrates that: (1) texture features from the gray level co-occurrence matrix (GLCM) significantly improved the classification of informal settlements, with the random forest classification model achieving a kappa coefficient above 0.77, an overall accuracy exceeding 0.89, and F1 scores above 0.90; (2) informal settlements exhibited two primary development patterns: gradual expansion near formal residential areas and dependence on natural resources such as farmland, forests, and water bodies; (3) economic vitality emerged as a critical factor in improving the living environment, while social, natural, and residential conditions remained relatively stable; (4) the proportion of highly suitable and moderately suitable areas increased from 65.62% to 65.92%, although the overall improvement in informal settlements remained slow. This study highlights the novel integration of multi-source remote sensing data with machine learning for precise spatial extraction and comprehensive habitat quality assessment, providing valuable insights into urban planning and sustainable development strategies.