Feature Space Model Research Articles

Explicit unsupervised feature selection based on structured graph and locally linear embedding

Numerous redundant and irrelevant features are usually contained in high-dimensional data whose presence, if ignored, can bring detrimental effects on the performance of data processing tasks. Unsupervised feature selection (UFS) as a trending topic in a great deal of domains has been greatly concerned, which can effectively filter out redundant and irrelevant features in the unlabeled data. The majority of existing UFS methods mainly focus on building the parameterized model in the original feature space or some certain low-dimensional feature subspace, which cannot fully capture the information of the target space. In this work, we propose a novel UFS method that unifies the explicit feature selective matrix and structured graph into a learning framework. An explicit selection matrix, whose scale, bound of the elements and structure are considered, is tailored for UFS. Furthermore, we take into account the local consistency so that the low dimensional manifold structures can be preserved to a better degree. Instead of using the pre-defined similarity matrix to construct graph, we make use of the way of learning to obtain the adaptive graph. Besides, for the purpose of exploiting the cluster structure of data, we impose a rank constraint on the graph. Notably, both the explicit selection matrix and intrinsic structure are learned in the target feature subspace, and thus the proposed methodology is more straightforward and effective. We present an effective and efficient algorithm for the resulting problem based on alternative optimization strategy, and establish its convergence and computational complexity analysis. The experimental results show that our proposed approach has improved by over 5.28% and 5.79% on average in terms of clustering accuracy (ACC) and Normalized Mutual Information (NMI) compared with comparison methods, which demonstrates its superiority.

Spatiotemporal evolution and driving mechanisms of desertification on the Mongolian Plateau

Desertification poses a severe ecological and environmental challenge in the Mongolian Plateau (MP). It is difficult to quantify desertification distribution using unified indicators in the entire MP, because of its complex physical geographic conditions and various climatic zones covered. To accurately address this challenge, the spatial distribution of desertification at a 30-m resolution from 1990 to 2020 were mapped in this study. The desertification potential occurrence zone was identified by using a moisture index on the MP firstly. The feature space model and five machine learning models were constructed to make the map based on Google Earth Engine and Landsat data. The spatiotemporal distribution of desertification were further analyzed, and the dominant drivers of desertification distribution and evolution were identified using Geodetector model. The results indicate that the potential occurrence area of desertification accounted for 83.88 % of the total land area. The gradient boosted tree model for desertification assessment has the best performance with the highest overall accuracy of 88.18 %. The year 2010 marked a pivotal transition from land degradation to land restoration in the MP. Between 2010 and 2020, desertified land continued to deteriorate extensively in the southern Mongolia, while Inner Mongolia, China, essentially entered a full recovery phase. Precipitation and land use emerged as the primary drivers of the spatial distribution of desertification on the Mongolian Plateau and Mongolia, with potential evapotranspiration and precipitation influencing the distribution of desertification in Inner Mongolia, China. Land use change was the primary driver of desertification evolution on the MP and Mongolia. This study constructs an indicator system and methodology suitable for desertification monitoring on the MP, addresses the lack of refined desertification data over a long time series, and provides scientific reference for decision-making support in combating desertification in this region, and other large arid and semi-arid areas in the world.

Monitoring soil salinity based on Sentinel-1/2 remote sensing parameters and two-dimensional space theory

Multi-spectral satellite remote sensing is widely used in soil salinity monitoring by virtue of its wide coverage, fast update and low cost, but it is susceptible to cloudy weather, limiting the accuracy and application conditions of the inversion, while SAR (synthetic aperture radar) is hardly affected by cloud with strong penetration ability and enables almost round-the-clock, climate-wide monitoring of soil salinity. Two-dimensional feature space model, being easily interpretive and widely used in soil salinity monitoring, requires only some easily extracted parameters and can reduce other influences on soil salinity inversion. Therefore, in order to take full advantages of the SAR and multi-spectral satellite, achieve monitoring of soil salinity in more climatic conditions and longer time, and further improve soil salinity monitoring accuracy, this paper combine SAR-Spectral data with two-dimensional space theory to construct two-dimensional models for monitoring soil salinity. In this study, two-dimensional space models are constructed using Sentinel-1 and Sentinel-2 feature parameters, and models are evaluated with R2 with measured soil salinity data of the Shahaoqu irrigation area in Hetao irrigation district, Inner Mongolia. The results show that it is feasible to use the parameters extracted from Sentinel-1 after H/a polarization for constructing two-dimensional feature space models for the soil salinity inversion(R2parameters of Sentinel-1 = 0.467,0.587,0.517), and the two-dimensional space models combining Sentinel-1/2 parameters are more accurate and stable in soil salinity monitoring than those using Sentinel-1 or Sentinel-2 alone((Rparameters of Sentinel-2 = 0.320,0.517,0.341),(R2parameters of Sentinel-1/2 = 0.507,0.545,0.562,0.512,0.557,0.568)).

A New Large-Scale Monitoring Index of Desertification Based on Kernel Normalized Difference Vegetation Index and Feature Space Model

As a new vegetation monitoring index, the KNDVI has certain advantages in characterizing the evolutionary process of regional desertification. However, there are few reports on desertification monitoring based on KNDVI and feature space models. In this study, seven feature parameters, including the kernel normalized difference vegetation index (KNDVI) and Albedo, were introduced to construct different models for desertification remote-sensing monitoring. The optimal desertification remote-sensing monitoring index model was determined with the measured data; then, the spatiotemporal evolution pattern of desertification in Gulang County from 2013 to 2023 was analyzed and revealed. The main conclusions were as follows: (1) Compared with the NDVI and MSAVI, the KNDVI showed more advantages in the characterization of the desertification evolution process. (2) The point–line pattern KNDVI-Albedo remote-sensing index model had the highest monitoring accuracy, reaching 94.93%, while the point–line pattern NDVI-TGSI remote-sensing monitoring index had the lowest accuracy of 54.38%. (3) From 2013 to 2023, the overall desertification situation in Gulang County showed a trend of improvement with a pattern of “firstly aggravation and then alleviation.” Additionally, the gravity center of desertification in Gulang County first shifted to the southeast and then to the northeast, indicating that the northeast’s aggravating rate of desertification was higher than in the southwest during the period. (4) From 2013 to 2023, the area of stable desertification in Gulang County was the largest, followed by the slightly weakened zone, and the most significant transition area was that of extreme desertification to severe desertification. The research results provide important decision support for the precise monitoring and governance of regional desertification.

Physics‐constrained symbolic model discovery for polyconvex incompressible hyperelastic materials

AbstractWe present a machine learning framework capable of consistently inferring mathematical expressions of hyperelastic energy functionals for incompressible materials from sparse experimental data and physical laws. To achieve this goal, we propose a polyconvex neural additive model (PNAM) that enables us to express the hyperelastic model in a learnable feature space while enforcing polyconvexity. An upshot of this feature space obtained via the PNAM is that (1) it is spanned by a set of univariate basis functions that can be re‐parametrized with a more complex mathematical form, and (2) the resultant elasticity model is guaranteed to fulfill the polyconvexity, which ensures that the acoustic tensor remains elliptic for any deformation. To further improve the interpretability, we use genetic programming to convert each univariate basis into a compact mathematical expression. The resultant multi‐variable mathematical models obtained from this proposed framework are not only more interpretable but are also proven to fulfill physical laws. By controlling the compactness of the learned symbolic form, the machine learning‐generated mathematical model also requires fewer arithmetic operations than its deep neural network counterparts during deployment. This latter attribute is crucial for scaling large‐scale simulations where the constitutive responses of every integration point must be updated within each incremental time step. We compare our proposed model discovery framework against other state‐of‐the‐art alternatives to assess the robustness and efficiency of the training algorithms and examine the trade‐off between interpretability, accuracy, and precision of the learned symbolic hyperelastic models obtained from different approaches. Our numerical results suggest that our approach extrapolates well outside the training data regime due to the precise incorporation of physics‐based knowledge.

A Novel Approach to Detecting the Salinization of the Yellow River Delta Using a Kernel Normalized Difference Vegetation Index and a Feature Space Model

Using the kernel normalized difference vegetation index (KNDVI) to monitor soil salinization has great advantages; however, approaches using KNDVI and a feature space model to monitor salinization have not yet been reported. In this study, the KNDVI, normalized difference vegetation index (NDVI), extended difference vegetation index (EDVI), green normalized difference vegetation index (TGDVI), modified soil-adjusted vegetation index (MSAVI), and salt index (SI) were used to establish five feature space monitoring indices for salinization. The spatio-temporal evolution pattern of soil salinization in the Yellow River Delta from 2000 to 2020 was analyzed based on the optimal monitoring index. The remote sensing monitoring index model based on KNDVI-SI’s point-to-point mode had the best applicability with R2 = 0.93, followed by EDVI-SI’s salinization monitoring index model with R2 = 0.90. From 2000 to 2020, soil salinization in the Yellow River Delta followed an exacerbating then improving trend. Soil salinization was more severe in the northern and eastern coastal areas of the Yellow River Delta. These results are conducive to salinization restoration and control in the Yellow River Delta.

SemiMAR: Semi-Supervised Learning for CT Metal Artifact Reduction.

Metal artifacts lead to CT imaging quality degradation. With the success of deep learning (DL) in medical imaging, a number of DL-based supervised methods have been developed for metal artifact reduction (MAR). Nonetheless, fully-supervised MAR methods based on simulated data do not perform well on clinical data due to the domain gap. Although this problem can be avoided in an unsupervised way to a certain degree, severe artifacts cannot be well suppressed in clinical practice. Recently, semi-supervised metal artifact reduction (MAR) methods have gained wide attention due to their ability in narrowing the domain gap and improving MAR performance in clinical data. However, these methods typically require large model sizes, posing challenges for optimization. To address this issue, we propose a novel semi-supervised MAR framework. In our framework, only the artifact-free parts are learned, and the artifacts are inferred by subtracting these clean parts from the metal-corrupted CT images. Our approach leverages a single generator to execute all complex transformations, thereby reducing the model's scale and preventing overlap between clean part and artifacts. To recover more tissue details, we distill the knowledge from the advanced dual-domain MAR network into our model in both image domain and latent feature space. The latent space constraint is achieved via contrastive learning. We also evaluate the impact of different generator architectures by investigating several mainstream deep learning-based MAR backbones. Our experiments demonstrate that the proposed method competes favorably with several state-of-the-art semi-supervised MAR techniques in both qualitative and quantitative aspects.

Desertification simulation using wavelet and box-jenkins time series analysis based on TGSI and albedo remote sensing indices

Desertification has been listed as one of the most critical global environmental issues, posing a significant threat to life, particularly in arid and semiarid regions. Therefore, gaining a comprehensive understanding of the present and future desertification trends becomes imperative. This study employs a feature space model, which effectively captures land surface changes related to desertification, enabling the extraction of pertinent information. Subsequently, time series models are used to determine the most accurate desertification simulation. Twenty-one ETM + sensor images were utilized to calculate the Topsoil Grain Size (TGSI) and Albedo remotely sensed indexes. Constructing the Albedo-TGSI feature space, the Desertification Degree Index (DDI) was extracted for each year. Different levels of desertification were identified by applying a natural break classification on the DDI values, and corresponding break values were obtained. The representative desertification degree for each year was determined by calculating the average of the minimum and maximum break values, resulting in the generation of five distinct time series for five desertification degrees. Different ARIMA models and wavelet transforms were selected to simulate the various desertification degrees based on the analysis of autocorrelation and partial autocorrelation functions and trial and error, respectively. The most suitable ARIMA models with the lowest errors were identified as follows: ARIMA (1,0,7) for severe desertification, ARIMA (0,1,6) for high desertification, ARIMA (0,0,7) for moderate desertification, and ARIMA (3,0,6) for non-desertification degrees. Among the various wavelet transform families tested, the Symlet family proved to be the most effective, except for the low desertification degree. The following wavelet transforms yielded the best results for each degree of desertification: Symlet3 for severe desertification, Symlet7 for high desertification, Symlet7 for moderate desertification, Daubechies 5 (db5) for low desertification, and Symlet7 for non-desertification degree simulations, all exhibiting the minimum error rates.

An evaluation of the response of vegetation greenness, moisture, fluorescence, and temperature-based remote sensing indicators to drought stress

Drought seriously threatens the food and ecological security of many countries in the world. Currently, various remote sensing indicators of vegetation status have been utilized in drought monitoring and warning. This study classified them into four types according to the biophysical characteristics of vegetation, i.e., temperature, moisture, fluorescence, and greenness-based indicators. However, their responses to drought stress are still less clear at present, which is not conducive to remote sensing of drought. Hence, a comprehensive evaluation of their responses to drought was conducted in this study. MODIS spectral reflectance and land surface temperature products were used to obtain temperature, moisture, and greenness-based indicators, while five solar-induced chlorophyll fluorescence datasets were used to obtain fluorescence-based indicators. Soil moisture anomaly, Standardized Precipitation Index, and Standardized Precipitation Evapotranspiration Index calculated by reanalysis and in-situ meteorological data were employed to indicate drought stress. The Chinese mainland was selected as the study area and the study period covers the year from 2000 to 2020 with months from May to September, the vegetation growing period. Evaluation results demonstrated that: 1) The responses of all indicators to drought stress were the most sensitive in arid areas, the second in transitional areas, and the last in humid areas; Monitoring or evaluation of drought with remote sensing vegetation indicators was more suitable for arid and transitional areas. 2) Among all indicators, the temperature-based indicator was found to be the most sensitive, the moisture-based indicator was in the second, the fluorescence-based and greenness-based indicators were in the third where the former presented a little more sensitive than the latter. 3) Vegetation temperature and moisture-based indicators should be integrated in the comprehensive indexes of drought rather than only integrate vegetation greenness-based indicators. Soil background effect in the remotely sensed vegetation indicators can be reduced with the feature space model.

Monitoring of deforestation events in the tropics using multidimensional features of Sentinel 1 radar data

Many countries and regions are currently developing new forest strategies to better address the challenges facing forest ecosystems. Timely and accurate monitoring of deforestation events is necessary to guide tropical forest management activities. Synthetic aperture radar (SAR) is less susceptible to weather conditions and plays an important role in high-frequency monitoring in cloudy regions. Currently, most SAR image-based deforestation identification uses manually supervised methods, which rely on high quality and sufficient samples. In this study, we aim to explore radar features that are sensitive to deforestation, focusing on developing a method (named 3DC) to automatically extract deforestation events using radar multidimensional features. First, we analyzed the effectiveness of radar backscatter intensity (BI), vegetation index (VI), and polarization feature (PF) in distinguishing deforestation areas from the background environment. Second, we selected the best-performing radar features to construct a multidimensional feature space model and used an unsupervised K-mean clustering method to identify deforestation areas. Finally, qualitative and quantitative methods were used to validate the performance of the proposed method. The results in Paraguay, Brazil, and Mexico showed that (1) the overall accuracy (OA) and F1 score (F1) of 3DC were 88.1–98.3% and 90.2–98.5%, respectively. (2) 3DC achieved similar accuracy to supervised methods without the need for samples. (3) 3DC matched well with Global Forest Change (GFC) maps and provided more detailed spatial information. Furthermore, we applied the 3DC to deforestation mapping in Paraguay and found that deforestation events occurred mainly in the second half of the year. To conclude, 3DC is a simple and efficient method for monitoring tropical deforestation events, which is expected to serve the restoration of forests after deforestation. This study is also valuable for the development and implementation of forest management policies in the tropics.

Reduced-Space Multistream Classification Based on Multiobjective Evolutionary Optimization

In traditional data stream mining, classification models are typically trained on labeled samples from a single source. However, in real-world scenarios, obtaining accurate labels is very hard and expensive, especially when multiple data streams are concurrently sampled from an environment or the same process. To address this issue, multistream classification is proposed, in which a data stream with biased labels (called the source stream) is leveraged to train a suitable model for prediction over another stream with unlabeled samples (called the target stream). Despite the growing research in this field, previous multistream classification methods are mostly designed for single source stream scenarios. However, various source streams contain diverse data distributions, providing more valuable information for building a more accurate model. In addition, previous works construct classification models in the original shared feature space, ignoring the effect of redundant or low-quality features on the classification performance. This may produce inefficient knowledge transfer across streams. In view of this, a reduced-space multistream classification based on multi-objective evolutionary optimization is proposed in this paper. First, a multi-objective evolutionary optimization is employed to seek the most valuable feature subset shared in the source and target domains, with the purpose of narrowing the distribution difference between source and target streams. Following that, a Gaussian Mixture Model-based weighting mechanism for source samples is presented. More especially, two drift adaptation methods are proposed to address asynchronous drift. Experimental results on benchmark datasets show that the proposed method outperforms other comparative methods on classification accuracy and G-mean.

Extraction of Rocky Desertification Information in the Karst Area Based on the Red-NIR-SWIR Spectral Feature Space

The complex topography, severe surface fragmentation and landscape heterogeneity of the karst region of southwest China make it extremely difficult to extract information on rocky desertification in the region. In order to overcome the disadvantages of the surface parameter-based feature space approach, which is difficult to construct and apply, this study uses the reflectance of Landsat 8 Operational Land Imager (OLI) in the red (Red), near-infrared (NIR) and shortwave infrared (SWIR) bands as the feature variables, and establishes a two-dimensional SWIR-NIR, Red-NIR and SWIR-Red reflectance spectral feature space. The three models of perpendicular rocky desertification index 1 (PRDI1), perpendicular rocky desertification index 2 (PRDI2) and perpendicular rocky desertification index 3 (PRDI3) were also constructed based on the variation of the degree of rocky desertification in each spectral feature space. The accuracy of the rocky desertification extracted by these three index models was verified and compared with the karst rocky desertification index (KRDI) and rocky desertification difference index (RSDDI), which are constructed based on the surface parameter feature space. The results show that: (1) The waveband reflectance-based feature space model provides a new method for large-scale rocky desertification information extraction, characterized by easy data acquisition, simple index calculation and good stability, and is conducive to the monitoring and quantitative analysis of rocky desertification in karst areas. (2) The overall accuracy and Kappa coefficient of PRDI1 are 0.829 and 0.784, respectively, both higher than other index models, showing the best applicability, accuracy and effectiveness in rocky desertification information extraction. (3) According to the results extracted from PRDI1, the total area of rocky desertification in Huaxi District of Guizhou province is 320.44 km2, with the more serious grades of rocky desertification, such as severe and moderate, mainly distributed in the southwestern, western and southeastern areas of Huaxi District. This study provides important information on the total area and spatial distribution of different degrees of rocky desertification in the study area, and these results can be used to support the local government’s ecological and environmental management decisions. The method proposed in this study is a scientific and necessary complement to the characteristic spatial methods based on different surface parameters, and can provide important methodological support for the rapid and efficient monitoring of karstic rocky desertification over large areas.

Synergy biochemical and physiological traits for rice heavy metal stress detection from Sentinel-2 images using signal–decomposition technique

The increasing pollution of soil with heavy metals has greatly damaged the ecological environment and endangered human health. Heavy metal stress affects not only biochemical constituents (i.e., chlorophyll) but also physiological functions (i.e., photosynthesis). We aimed to detect heavy metal stress in rice by integrating these two aspects. We selected Zhuzhou, Hunan Province as the study area and collected Sentinel-2 images from 2019 to 2021 and situ-measured data. First, two indicators were computed from Sentinel-2 images characterizing biochemical components and physiological functions, respectively. Second, the original time series were decomposed using the complete ensemble empirical mode decomposition with adaptive noise. Finally, a time-spectrum feature space (TSFS) model based on signal decomposition in critical periods sensitive to heavy metal stress was developed to determine the level of heavy metal stress. It showed that, regardless of individuals years, the overall accuracy and kappa coefficients of the TSFS model were >85 % and 0.8, and compared with a single indicator, combining two indicators with the signal-decomposition method significantly improved the accuracy of stress detection. This demonstrates that the integration of biochemical components with physiological functions and signal decomposition techniques can provide an important reference for crop heavy metal stress monitoring.

Spatial-temporal pattern of desertification in the Selenge River Basin of Mongolia from 1990 to 2020

Land degradation is the most serious environmental challenge in the Mongolian Plateau, an important arid and semiarid region east of the Eurasian continent. The Selenge River Basin is not only the main catchment area of Baikal Lake, the largest fresh water lake, but also the main concentration area of agriculture and animal husbandry in Mongolia. Under the common influence of global warming and human activities, desertification has become more prominent in this basin, threatening the ecological security and sustainable development of the Mongolian Plateau. In this study, we selected NDVI, Modified Soil Adjusted Vegetation Index, topsoil grain size index and Albedo as feature space indicators, and retrieved the desertification process from 1990 to 2020 in the Selenge River Basin of Mongolia based on a novel feature space monitoring index. A 30-m resolution desertification map of the Selenge River Basin was retrieved based on optimal feature space models for 1990, 1995, 2000, 2005, 2010, 2015, and 2020. Then, the spatial-temporal dynamic changes and driving mechanism of desertification. The results show that: 1) Compared with the other four feature space models, the point-to-line Albedo-MSAVI feature space model has the highest recognition accuracy of 84.89% for desertification in the basin. 2) The desertification level of the Selenge River basin is mainly low and medium on the whole, the high desertification is mainly located in BULGAN and HOVSGOL provinces in the middle-upper reaches of the basin, and the severe desertification is mainly located in TOV province and Ulaanbaatar in the middle-lower reaches of the basin. 3) From 1990–2020, desertification degree in the Selenge River Basin has further deteriorated, and the area of high and serve desertified land has expanded significantly. Within the stage, 1990–2015 was a period of rapid increase in desertification. However, from 2015–2020, recovery takes the dominant position. The regions with high conversion frequency of desertification degree are mainly concentrated in the central and southeastern of the Selenge River basin. The joint effects of large fluctuations in temperature, overgrazing and population migration aggravate the desertification degree in this region. The research results can provide the desertification retrieving method recommendation and land degradation nutrition measures decision support in the Selenge River Basin and the whole Mongolian Plateau.

Domain generalization improves end-to-end object detection for real-time surgical tool detection.

Computer assistance for endoscopic surgery depends on knowledge about the contents in an endoscopic scene. An important step of analysing the video contents is real-time surgical tool detection. Most methods for tool detection nevertheless depend on multi-step algorithms building upon prior knowledge like anchor boxes or non-maximum suppression which ultimately decrease performance. A real-world difficulty encountered by learning-based methods are limited datasets. Training a neural network on data matching a specific distribution (e.g. from a single hospital or showing a specific type of surgery) can result in a lack of generalization. In this paper, we propose the application of a transformer based architecture for end-to-end tool detection. This architecture promises state-of-the-art accuracy while decreasing the complexity resulting in improved run-time performance. To improve the lack of cross-domain generalization due to limited datasets, we enhance the architecture with a latent feature space via variational encoding to capture common intra-domain information. This feature space models the linear dependencies between domains by constraining their rank. The trained neural networks show a distinct improvement on out-of-domain data indicating better generalization to unseen domains. Inference with the end-to-end architecture can be performed at up to 138 frames per second (FPS) achieving a speedup in comparison to older approaches. Experimental results on three representative datasets demonstrate the performance of the method. We also show that our approach leads to better domain generalization.

A novel remote sensing monitoring index of salinization based on three-dimensional feature space model and its application in the Yellow River Delta of China

Previous studies were mostly conducted based on two-dimensional feature space to monitor salinization, while studies on dense long-term salinization monitoring based on three-dimensional feature space have not been reported. Based on Landsat TM/ETM+/OLI images and three-dimensional feature space method, this study introduced six typical salinization surface parameters, including NDVI, salinity index, MSAVI, surface albedo, iron oxide index, wetness index to construct eight different feature space monitoring index. The optimal soil salinization monitoring index model was proposed base on field observed data and then the evolution process of salinization in Yellow River Delta (YRD) were analyzed and revealed during 1984–2022. The salinization monitoring index model of MSAVI-Albedo-IFe2O3 feature space had the highest accuracy with R2 = 0.93 and RMSE = 0.678g/kg. The spatial distribution of salinization in YRD showed an increasing trend from inland southwest to coastal northeast and the salinization intensity showed an increasing trend during 1984–2022 due to the implements of agricultural measures such as planting salt-tolerant crops, microbial remediation and fertility improvement. The rate of salinization deterioration in the northeast part was greater than others. Zones of salinization improvement were mainly located in cultivated land of the southwest parts.

The dominant driving factors of rocky desertification and their variations in typical mountainous karst areas of southwest China in the context of global change

Significant differences exist in the dominant driving factors of rocky desertification evolution in varying temporal periods of mountainous karst areas in southwest China. However, previous studies were primarily conducted in specific periods. In this study, taking Bijie city (one of the areas most seriously affected by rocky desertification in southwest China) as an example, we analyzed the spatio-temporal evolution pattern of rocky desertification in Bijie city in recent 35 years by introducing the feature space model and the gravity centre model, and then clarified the dominant driving factors of rocky desertification in the study area in varying temporal periods based on GeoDetector. The findings were as follows: (1) the proposed point-line BLI-NDVI feature space model has higher applicability for rocky desertification monitoring with the overall inversion accuracy of 92.0 %; (2) during 1985–2020, there was an overall decreasing trend (weakening-aggravating-weakening) of the rocky desertification; (3) the spatial distributions and migration trajectory of gravity centre of rocky desertification indicated that the rocky desertification degree in the western parts was more severely than that in the eastern regions; (4) during 1985–2020, the land use and POP density have become the dominant factors of rocky desertification instead of vegetation coverage and temperature.

A feature augmentation-based method for constructing generative-discriminative hybrid models

From the perspective of probability framework, the generative model first learns the joint probability distribution from the data and then calculates the conditional probability distribution with a faster convergence speed. However, the discriminative model learns the conditional probability distribution directly from the data, thus often demonstrating higher accuracy. As an effective combination of the generative and discriminative models, the generative-discriminative hybrid model integrates their advantages. However, the existing methods must divide the original features into two independent feature spaces to train the two models. Feature division not only increases the time complexity of the model but also weakens the expression ability of the original feature space. To solve this problem, this paper proposes a feature augmentation-based method for constructing the generative-discriminative hybrid model. First, this novel method uses the generative model to learn the conditional probability distribution. Then, it augments the learned conditional probability distribution as new features into the original feature space. Finally, it trains the discriminative model in the augmented feature space and predicts the final classification result. The new method offers several advantages, including using feature augmentation to mix the models, not requiring feature division, exhibiting low time complexity, and enhancing the expression ability of the original feature space. The experimental results on 36 classical UCI benchmark datasets show that the new method is not only effective and universal but also follows bias-variance trade-off.

Spatial and Temporal Pattern Changes and Driving Forces: Analysis of Salinization in the Yellow River Delta from 2015 to 2020

China's Yellow River Delta represents a typical area with moist semi-humid soil salinization, and its salinization has seriously affected the sustainable use of local resources. The use of remote sensing technology to understand changes in the spatial and temporal patterns of salinization is key to combating regional land degradation. In this study, a feature space model was constructed for remote sensing and monitoring land salinization using Landsat 8 OIL multi-spectral images. The feature parameters were paired to construct a feature space model; a total of eight feature space models were obtained. An accuracy analysis was conducted by combining salt-loving vegetation data with measured data, and the model demonstrating the highest accuracy was selected to develop salinization inversion maps for 2015 and 2020. The results showed that: (1) The total salinization area of the Yellow River Delta displayed a slight upward trend, increasing from 4244 km2 in 2015 to 4629 km2 in 2020. However, the area's salting degree reduced substantially, and the areas of saline soil and severe salinization were reduced in size; (2) The areas with reduced salinization severity were mainly concentrated in areas surrounding cities, and primarily comprised wetlands and some regions around the Bohai Sea; (3) Numerous factors such as the implementation of the “Bohai Granary” cultivation engagement plan, increase in human activities to greening local residential living environments, and seawater intrusion caused by the reduction of sediment contents have impacted the distribution of salinization areas in the Yellow River Delta; (4) The characteristic space method of salinization monitoring has better applicability and can be promoted in humid-sub humid regions.

A novel feature space monitoring index of salinisation in the Yellow River Delta based on SENTINEL‐2B MSI images

AbstractMost of previous studies utilized the surface parameters from LANDSAT images to construct the feature space monitoring index model of salinisation (salinization), and a few studies that have combined the feature space model with SENTINEL‐2B MSI images have been reported. In addition, the red edge index derived from SENTINEL‐2B MSI images can provide more detailed information to indicate the vegetation condition when monitoring salinized land ecosystems. Based on SENTINEL‐2B MSI images, this paper introduces seven typical parameters, namely NDVI, MSAVI, SI, Albedo, NDre1, NDre2, and NDre3 (red edge index) to construct two category features space models (point‐to‐line type and point‐to‐point type), and then, a novel salinisation monitoring index for use in the Yellow River Delta (YRD). Our main conclusions showed that: (1) the monitoring index model based on SENTINEL‐2B MSI images and a feature space model has high applicability for the salinisation monitoring in the YRD, with an average precision of R2 = 0.8499; (2) the point‐to‐point monitoring index of soil salinisation based on the NDre1‐SI feature space model has the best inversion accuracy of R2 = 0.9305 and RMSE = 0.9926; (3) the red edge index can better indicate the state and evolution process of soil salinisation. The salinisation monitoring models that included the red edge indexes have higher inversion accuracy with an average value of R2 = 0.8650; (4) the soil salinisation in the YRD was more serious in its eastern and northeastern regions than other parts. The results provide a new technical and methodological approach for the prevention and treatment of regional salinisation.