Snow Classification Research Articles

Strategy-Performance Relationship Through Organizational Resilience

This paper aims to analyze the 5strategy-performance relationship through the mediation of organization resilience in tourism SMEs. Primary data was collected from 760 managers of tourism SMEs in the top five tourism destinations in Khyber Pakhtunkhwa (KP) province of Pakistan using multistage cluster sampling through researcher-administrated questionnaires. Miles and Snow's classification of SMEs' strategic orientations as Prospector, Analyzer, Defender and Reactors was used. AMOS-SEM was applied to analyze the strategy-performance relationship and mediation of organizational resilience. The findings showed that strategy has a significant impact on organizational resilience and organizational performance. Organizational resilience mediates the strategy-performance relationship. Prospectors displayed the highest performance, followed by analyzers and defenders, while the performance of reactors was the lowest. This was cross-sectional research conducted in tourism SMEs in the Pakistani context. The sector is time-specific; therefore, a longitudinal study can be undertaken in national and international contexts. The government’s support in developing organizational resilience is rudimentary in the tourism sector due to its high vulnerability to environmental jolts. This empirical research contributed to the least-researched sector of tourism SMEs in Pakistan operating in distinct segments of tourism SMEs.

Automated evaluation of snow and ice conditions on airport pavement: A study based on image recognition and point cloud reconstruction

Automated evaluation of snow and ice conditions on airport pavement: A study based on image recognition and point cloud reconstruction

Airborne lidar intensity correction for mapping snow cover extent and effective grain size in mountainous terrain

ABSTRACT Differentially mapping snow depth in mountain watersheds from airborne laser altimetry is a valuable hydrologic technique that has seen an expanded use in recent years. Additionally, lidar systems also record the strength of the returned light pulse (i.e. intensity), which can be used to characterize snow surface properties. For near-infrared lidar systems, return intensity is relatively high over snow and inversely related to the effective grain size, a primary control on snow albedo. Raw intensity is also sensitive to laser range and incidence angle, however, and requires a correction for snow property retrieval that is especially pertinent in mountainous terrain. Here, we describe a workflow to correct the intensity using the plane trajectory, lidar scan angle, and lidar-derived topography. As a proof of concept for snow retrievals, we apply the workflow to an airborne 1064 nm lidar flight over a snow-covered mountain basin in the Colorado Rockies. Corrected intensity was empirically related to reflectance before delineating snow extent and retrieving grain size. Relative to the traditional snow classification derived from optical imagery, the lidar-derived snow extent covered 5.4% more area due to the fine resolution point cloud and absence of shadows common in optical imagery. The lidar-derived grain size retrievals had a MAE of 32 µm compared to those from field spectroscopy, which translated to a 1% error in snow albedo. We found high incidence angles yielded an overcorrection in intensity that introduced a high bias in the grain size distribution and, therefore, suggest using an incidence angle threshold (40°). Developing methods specifically for quantitative snow surface property retrievals from lidar intensity is timely and relevant as aerial lidar is increasingly being used to map snow depth for hydrologic and cryospheric studies.

Monitoring Snow Cover in Typical Forested Areas Using a Multi-Spectral Feature Fusion Approach

Accurate snow cover monitoring is greatly significant for research on the hydrology model and regional climate variation, especially in Northeast China where forests cover almost forty percent of the total area. However, effectively monitoring snow cover under the forest canopy is still challenging with either in situ or remote sensing observations. The global SNOWMAP algorithm pertinent to the fixed normalized difference snow index (NDSI) threshold is, therefore, no longer applicable in a typical forested region of Northeast China. In order to achieve the goal of improving the accuracy of monitoring snow cover in areas with forest, utilizing MOD09GA and MOD13A1 products, a new approach of snow mapping was developed in this study, and it exploits the fusion and coupling of spectral features by integrating and analyzing the normalized difference forest snow index (NDFSI), the normalized difference vegetation index (NDVI), and the NDSI index. Then, Landsat 8 OLI images of high resolution were used to evaluate snow cover mapping precision. The experimental results indicated that the NDFSI index combined with the NDVI index showed great potential for extracting the snow cover distribution in forested regions. Compared with the snow distribution obtained from the Landsat 8 images, the average bias and FAR (false alarm ratio) values of snow cover mapping obtained by this algorithm were 1.23 and 13.54%, which were reduced by 1.98 and 29.36%, respectively. The overall accuracy of 81.31% was reached, which is improved by 20.19%. Thus, the snow classification scheme combining multiple spectral features from MODIS data works effectively in improving the precision of automatic snow cover mapping in typical forested areas of Northeast China, which provides essential support and significant perspectives for the next step of establishing a runoff model and rationally regulating forest water resources.

Comparative plastome analysis and taxonomic classification of snow lotus species (Saussurea, Asteraceae) in Central Asia and Southern Siberia

Four species of Saussurea, namely S. involucrata, S. orgaadayi, S. bogedaensis, and S. dorogostaiskii, are known as the “snow lotus,” which are used as traditional medicines in China (Xinjiang), Kyrgyzstan, Mongolia, and Russia (Southern Siberia). These species are threatened globally, because of illegal harvesting and climate change. Furthermore, the taxonomic classification and identification of these threatened species remain unclear owing to limited research. The misidentification of medicinal species can sometimes be harmful to health. Therefore, the phylogenetic and genomic features of these species need to be confirmed. In this study, we sequenced five complete chloroplast genomes and seven nuclear ITS regions of four snow lotus species and other Saussurea species. We further explored their genetic variety, selective pressure at the sequence level, and phylogenetic relationships using the chloroplast genome, nuclear partial DNA sequences, and morphological features. Plastome of the snow lotus species has a conserved structure and gene content similar to most Saussurea species. Two intergenic regions (ndhJ–ndhK and ndhD-psaC) show significantly high diversity among chloroplast regions. Thus, ITS and these markers are suitable for identifying snow lotus species. In addition, we characterized 43 simple sequence repeats that may be useful in future population genetic studies. Analysis of the selection signatures identified three genes (rpoA, ndhB, and ycf2) that underwent positive selection. These genes may play important roles in the adaptation of the snow lotus species to alpine environments. S. dorogostaiskii is close to S. baicalensis and exhibits slightly different adaptation from others. The taxonomic position of the snow lotus species, confirmed by morphological and molecular evidence, is as follows: (i) S. involucrata has been excluded from the Mongolian flora due to misidentification as S. orgaadayi or S. bogedaensis for a long time; (ii) S. dorogostaiskii belongs to section Pycnocephala subgenus Saussurea, whereas other the snow lotus species belong to section Amphilaena subgenus Amphilaena; and (iii) S. krasnoborovii is synonymous of S. dorogostaiskii. This study clarified the speciation and lineage diversification of the snow lotus species in Central Asia and Southern Siberia.

Global Snow Seasonality Regimes from Satellite Records of Snow Cover

Abstract Snow cover provides distinct seasonal controls on the exchange of energy between Earth’s surface and atmosphere, and on hydrologic cycling, and holds considerable importance to communities and ecosystems worldwide. In this work, we tackle a comprehensive review of existing snow classification approaches and the development of new globally applicable snow cover–based rules for delineating snow seasonality classes. Snow classification rules are defined using machine learning approaches, which are then applied to the 22-yr record of snow cover (2000–22) from the Moderate Resolution Imaging Spectroradiometer (MODIS) on a 0.01° global grid. For the MODIS period of record, we find the global land surface can be effectively partitioned into five snow seasonality classes: no snow, ephemeral, transitional, seasonal, and perennial snow regimes which on average cover extents of approximately 76 (52% of global land areas), 19 (13%), 16 (11%), 18 (13%), and 16 million km2 (11%), respectively. Using the multidecadal dataset, we explore changes within snow regimes and find significant increases in the areal extent of no snow (approximately +70 000 km2 yr−1) as well as apparent losses in perennial (−3600 km2 yr−1) and seasonal snow regime coverage (−38 000 km2 yr−1). The resulting classification maps have strong agreement with in situ snow depth observations and present similar patterns to existing snow and climate classifications with notable discrepancies in cold arid regions. The framework’s ability to accurately capture variations in snow persistence, snow accumulation, and melt cycling is shown, providing a reference to the current state of global snow seasonality. Significance Statement Following a review of existing snow classification approaches, this study focuses on improving our understanding of snow-cover variability on Earth by using satellite-based observations of snow-covered area to derive a new snow classification. Satellite observations provide the best means of measuring snow cover at a global scale, helping to identify regions where its influence on energy, water, and climate is changing. The results are compared to existing climate and snow summaries, ground observations of snow depth, and include trends in snow cover over recent decades (2000–22). The resulting datasets are also made available to the broader scientific community.

DSRSS-Net: Improved-Resolution Snow Cover Mapping from FY-4A Satellite Images Using the Dual-Branch Super-Resolution Semantic Segmentation Network

The Qinghai–Tibet Plateau is one of the regions with the highest snow accumulation in China. Although the Fengyun-4A (FY4A) satellite is capable of monitoring snow-covered areas in real time and on a wide scale at high temporal resolution, its spatial resolution is low. In this study, the Qinghai–Tibet Plateau, which has a harsh climate with few meteorological stations, was selected as the study area. We propose a deep learning model called the Dual-Branch Super-Resolution Semantic Segmentation Network (DSRSS-Net), in which one branch focuses with super resolution to obtain high-resolution snow distributions and the other branch carries out semantic segmentation to achieve accurate snow recognition. An edge enhancement module and coordinated attention mechanism were introduced into the network to improve the classification performance and edge segmentation effect for cloud versus snow. Multi-task loss is also used for optimization, including feature affinity loss and edge loss, to obtain fine structural information and improve edge segmentation. The 1 km resolution image obtained by coupling bands 1, 2, and 3; the 2 km resolution image obtained by coupling bands 4, 5, and 6; and the 500 m resolution image for a single channel, band 2, were inputted into the model for training. The accuracy of this model was verified using ground-based meteorological station data. Snow classification accuracy, false detection rate, and total classification accuracy were compared with the MOD10A1 snow product. The results show that, compared with MOD10A1, the snow classification accuracy and the average total accuracy of DSRSS-Net improved by 4.45% and 5.1%, respectively. The proposed method effectively reduces the misidentification of clouds and snow, has higher classification accuracy, and effectively improves the spatial resolution of FY-4A satellite snow cover products.

Automatic snow type classification of snow micropenetrometer profiles with machine learning algorithms

Abstract. Snow-layer segmentation and classification are essential diagnostic tasks for various cryospheric applications. The SnowMicroPen (SMP) measures the snowpack's penetration force at submillimeter intervals in snow depth. The resulting depth–force profile can be parameterized for density and specific surface area. However, no information on traditional snow types is currently extracted automatically. The labeling of snow types is a time-intensive task that requires practice and becomes infeasible for large datasets. Previous work showed that automated segmentation and classification is, in theory, possible but cannot be applied to data straight from the field or needs additional time-costly information, such as from classified snow pits. We evaluate how well machine learning models can automatically segment and classify SMP profiles to address this gap. We trained 14 models, among them semi-supervised models and artificial neural networks (ANNs), on the MOSAiC SMP dataset, an extensive collection of snow profiles on Arctic sea ice. SMP profiles can be successfully segmented and classified into snow classes based solely on the SMP's signal. The model comparison provided in this study enables SMP users to choose a suitable model for their task and dataset. The findings presented will facilitate and accelerate snow type identification through SMP profiles. Anyone can access the tools and models needed to automate snow type identification via the software repository “snowdragon”. Overall, snowdragon creates a link between traditional snow classification and high-resolution force–depth profiles. Traditional snow profile observations can be compared to SMP profiles with such a tool.

Evaluating MODIS snow products using an extensive wildlife camera network

Snow covers a maximum of 47 million km2 of Earth's northern hemisphere each winter and is an important component of the planet's energy balance, hydrology cycles, and ecosystems. Monitoring regional and global snow cover has increased in urgency in recent years due to warming temperatures and declines in snow cover extent. Optical satellite instruments provide large-scale observations of snow cover, but cloud cover and dense forest canopy can reduce accuracy in mapping snow cover. Remote camera networks deployed for wildlife monitoring operate below cloud cover and in forests, representing a virtually untapped source of snow cover observations to supplement satellite observations. Using images from 1181 wildlife cameras deployed by the Norwegian Institute for Nature Research (NINA), we compared snow cover extracted from camera images to Moderate Resolution Imaging Spectroradiometer (MODIS) snow cover products during winter months of 2018–2020. Ordinal snow classifications (scale = 0–4) from cameras were closely related to normalized difference snow index (NDSI) values from the MODIS Terra Snow Cover Daily L3 Global 500 m (MOD10A1) Collection 6 product (R2 = 0.70). Tree canopy cover, the normalized difference vegetation index (NDVI), and image color mode influenced agreement between camera images and MOD10A1 NDSI values. For MOD10A1F, MOD10A1's corresponding cloud-gap filled product, agreement with cloud-gap filled values decreased from 78.5% to 56.4% in the first three days of cloudy periods and stabilized thereafter. Using our camera data as validation, we derived a threshold to create daily binary maps of snow cover from the MOD10A1 product. The threshold corresponding to snow presence was an NDSI value of 40.50, which closely matched a previously defined global binary threshold of 40 using the MOD10A2 8-day product. These analyses demonstrate the utility of camera trap networks for validation of snow cover products from satellite remote sensing, as well as their potential to identify sources of inaccuracy.

Dual Gradient Based Snow Attentive Desnowing

In this paper, we propose a novel dual gradient-based desnowing algorithm that can accurately remove snow from a scene by characterizing snow particles. To localize snow in an image, we present a gradient-based snow activation map that can be estimated using snow classification. To recognize various patterns in the shapes and trajectories of snow particles, we introduce a gradient-based snow edge map. Using these two gradients, we estimate an accurate snow attention mask that is subsequently used for snow removal. In addition, we propose a translucency-aware context restoration network to handle various degrees of snow transparency and thus, prevent our method from losing the image context information during desnowing. Experimental results demonstrate that the proposed method considerably outperforms other state-of-the-art desnowing algorithms quantitatively and qualitatively.

Statistical Analysis of Sparkle in Snow Images

Sparkle from snow is a common phenomenon in Nature but not well studied in the literature. We perform a statistical study on digital snow images captured in-situ to analyze sparkle events by using contrast and density of sparkle spots descriptors. The method for measuring sparkles by Ferrero et al. is adapted, tested, and verified to the case of snow. The dataset is divided into three categories representing the type of snow acquired: dense snow, fresh snow, and old snow. Our analysis highlights the link between the sparkle of snow, the nature of snow and its grain structure. Sparkle could thus be a feature used for snow classification.

Near-infrared LED system to recognize road surface conditions for autonomous vehicles

Abstract. The driving safety of autonomous vehicles will strongly depend on their ability to recognize road surface conditions such as dry, wet, snowy and icy road. Currently, the existing investigations to detect road surface conditions still have limitations in daytime and nighttime conditions. The objective of this paper is to propose and develop a new system with three near-infrared (NIR) LED sources. This choice is based on the advantages of LED sources over laser diodes. They are less sensitive to temperature and have lower costs. Considering these advantages, the feasibility of the LED system to recognize road surface conditions is investigated. For this, the appropriate wavelengths of the LED tri-wavelength source are first computed from experimental data taking into account the specific LED spectral shape. In addition, the effect of the spectral bandwidth of the LED sources on the system performance is theoretically studied. Finally, the NIR LED system with the LED sources at 970, 1450 and 1550 nm is experimentally tested and validated with an incident angle from 78.7 to 86.2∘. According to the results of the experiments, the accuracy of the classification of snow, wet and water can reach 97 %, while the accuracy of the dry and wet road surface conditions is respectively 73 % and 68 %.

The Collection of Hyperspectral Measurements on Snow and Ice Covers in Polar Regions (SISpec 2.0)

The data value of hyperspectral measurements on ice and snow cover is strongly impacted by the availability of data services, where spectral libraries are integrated to detailed descriptions of the observed surface cover. For snow and ice cover, we present an updated version of the Snow/Ice Spectral Archive (SISpec 2.0), which has been integrated into a web portal characterized by different functionalities. The adopted metadata scheme features basic geographic data, information about the acquisition setup, and parameters describing the different surface types. While the implementation of the IACS Classification of Seasonal Snow on the Ground is the core component for snow cover, ice cover is approached using different parameters associated with its surface roughness and location. The web portal is not only a visualization tool, but also supports interoperability functionalities, providing data in the NetCDF file format. The availability of these functionalities sets the foundation for sharing a novel platform with the community and is an interesting tool for calibrating and validating data and models.

Quantitative approach for snowy feature detection using polarimetric analysis

Synthetic aperture radar is an advanced remote sensing and imaging radar. It plays vital role in acquiring high resolution images of earth surface. The capturing of images by synthetic aperture radar is done in any season immaterial of weather conditions. This paper gives the details of the basic feature extraction for the snow images. The two sample images are analyzed to know the feature details of the object under consideration. Analytical details of variation in entropy and the polarization were considered. The scattering mechanism involved in the snow area is analyzed. The details of snow classification based on its layered structure along with its physical nature like moisture involved are presented. The results indicate a high value of entropy of 0.94 for the snow image. The reason for high entropy is because of more surface uniformity in the snow images. The flat surface structured snow basically exhibits the surface scattering mechanism.

Joint Characterization of the Cryospheric Spectral Feature Space

Multispectral and hyperspectral feature spaces are useful for a variety of remote sensing applications ranging from spectral mixture modeling to discrete thematic classification. In many of these applications, models are used to project the higher dimensional continuum of reflectances (or radiances) onto lower dimensional mappings of the image target’s physical properties or categorical composition. In such cases, characterization of the feature space dimensionality, geometry and topology can provide fundamental guidance for effective model design. Utility of this characterization, however, hinges on identification of appropriate basis vectors for the feature space. The objective of this study is to compare and contrast two fundamentally different approaches for identifying feature space basis vectors via dimensionality reduction. In so doing, we illustrate how these two approaches can be combined to render a joint characterization that reveals spectral properties not apparent using either approach alone. We use a diverse collection of AVIRIS-NG reflectance spectra of ice and snow to illustrate the utility of the joint characterization to facilitate both modeling and classification of snow and ice reflectance. Joint characterization is also shown to assist with interpretation of physical properties inferred from the spectra. Spectral feature spaces combining principal components (PCs) and t-distributed Stochastic Neighbor Embeddings (t-SNEs) provide both physically interpretable dimensions representing the global structure of cryospheric reflectance properties as well as local manifold structures revealing clustering not resolved within the global continuum. The joint characterization reveals distinct continua for snow-firn gradients on different parts of the Greenland Ice Sheet and multiple clusters of ice reflectance properties common to both glacier and sea ice in different locations. The clustering revealed in the t-SNE feature spaces, and extended to the joint characterization, distinguishes subtle differences in spectral curvature specific to different spatial locations within the snow accumulation zone, as well as BRDF effects related to view geometry. The ability of the PC + t-SNE joint characterization to produce a physically interpretable spectral feature space revealing global topology while preserving local manifold structures for cryospheric hyperspectra suggests that this type of characterization might be extended to the much higher dimensional hyperspectral feature space of all terrestrial land cover.

Quantitative Precipitation Estimation over Antarctica Using Different Ze-SR Relationships Based on Snowfall Classification Combining Ground Observations

Snow plays a crucial role in the hydrological cycle and energy budget of the Earth, and remote sensing instruments with the necessary spatial coverage, resolution, and temporal sampling are essential for snowfall monitoring. Among such instruments, ground-radars have scanning capability and a resolution that make it possible to obtain a 3D structure of precipitating systems or vertical profiles when used in profiling mode. Radars from space have a lower spatial resolution, but they provide a global view. However, radar-based quantitative estimates of solid precipitation are still a challenge due to the variability of the microphysical, geometrical, and electrical features of snow particles. Estimations of snowfall rate are usually accomplished using empirical, long-term relationships between the equivalent radar reflectivity factor (Ze) and the liquid-equivalent snowfall rate (SR). Nevertheless, very few relationships take advantage of the direct estimation of the microphysical characteristics of snowflakes. In this work, we used a K-band vertically pointing radar collocated with a laser disdrometer to develop Ze-SR relationships as a function of snow classification. The two instruments were located at the Italian Antarctic Station Mario Zucchelli. The K-band radar probes the low-level atmospheric layers, recording power spectra at 32 vertical range gates. It was set at a high vertical resolution (35 m), with the first trusted range gate at a height of only 100 m. The disdrometer was able to provide information on the particle size distribution just below the trusted radar gate. Snow particles were classified into six categories (aggregate, dendrite aggregate, plate aggregate, pristine, dendrite pristine, plate pristine). The method was applied to the snowfall events of the Antarctic summer seasons of 2018–2019 and 2019–2020, with a total of 23,566 min of precipitation, 15.3% of which was recognized as showing aggregate features, 33.3% dendrite aggregate, 7.3% plates aggregate, 12.5% pristine, 24% dendrite pristine, and 7.6% plate pristine. Applying the appropriate Ze-SR relationship in each snow category, we calculated a total of 87 mm water equivalent, differing from the total found by applying a unique Ze-SR. Our estimates were also benchmarked against a colocated Alter-shielded weighing gauge, resulting in a difference of 3% in the analyzed periods.

Revisiting the Global Seasonal Snow Classification: An Updated Dataset for Earth System Applications

AbstractTwenty-five years ago, we published a global seasonal snow classification now widely used in snow research, physical geography, and as a mission planning tool for remote sensing snow studies. Performing the classification requires global datasets of air temperature, precipitation, and land-cover. When introduced in 1995, the finest resolution global datasets of these variables were on a 0.5° × 0.5° latitude-longitude grid (approximately 50 km). Here we revisit the snow classification system and, using new datasets and methods, present a revised classification on a 10-arcsecond × 10-arcsecond latitude-longitude grid (approximately 300 m). We downscaled 0.1° × 0.1° latitude-longitude (approximately 10 km) gridded meteorological climatologies (1981-2019, European Centre for Medium-Range Weather Forecasts [ECMWF] ReAnalysis, 5thGeneration Land [ERA5-Land]) using MicroMet, a spatially distributed, high-resolution, micro-meteorological model. The resulting air temperature and precipitation datasets were combined with European Space Agency (ESA) Climate Change Initiative (CCI) GlobCover land-cover data (as a surrogate for wind speed) to produce the updated classification, which we have applied to all of Earth’s terrestrial areas. We describe this new, high-resolution snow classification dataset, highlight the improvements added to the classification system since its inception, and discuss the utility of the climatological snow classes at this much higher resolution. The snow class dataset (Global Seasonal-Snow Classification 2.0) and the tools used to develop the data are publicly available online at the National Snow and Ice Data Center (NSIDC).

High-resolution CubeSat imagery and machine learning for detailed snow-covered area

Snow cover affects a diverse array of physical, ecological, and societal systems. As such, the development of optical remote sensing techniques to measure snow-covered area (SCA) has enabled progress in a wide variety of research domains. However, in many cases, the spatial and temporal resolutions of currently available remotely sensed SCA products are insufficient to capture SCA evolution at spatial and temporal resolutions relevant to the study of fine-scale spatially heterogeneous phenomena. We developed a convolutional neural network-based method to identify snow covered area using the ~3 m, 4-band PlanetScope optical satellite image dataset with ~daily, near-global coverage. By comparing our model performance to snow extent derived from high-resolution airborne lidar differential depth measurements and satellite platforms in two North American sites (Sierra Nevada, CA, USA and Rocky Mountains, CO, USA), we show that these emerging image archives have great potential to accurately observe snow-covered area at high spatial and temporal resolutions despite limited radiometric bandwidth and band placement. We achieve average snow classification F-Scores of 0.73 in our training basin and 0.67 in a climatically-distinct out-of-sample basin, suggesting opportunities for model transferability. We also evaluate the performance of these data in forested regions, suggesting avenues for further research. The unparalleled spatial and temporal coverage of CubeSat imagery offers an excellent opportunity for satellite remote sensing of snow, with real implications for ecological and water resource applications. • Daily, 3-m snow covered area derived from Planet Labs Inc. “PlanetScope” data. • Model-based neural network approach enabled by airborne lidar. • Derived snow covered area compares well to Sentinel-2 and Landsat-8. • Good performance in both California and Colorado, USA, suggesting transferability. • Limited performance around tree canopies.

A framework for classification of snow- and icephobicity

For decades, there has been no proper thermodynamic definition of icephobicity. Some claim icephobic properties at an adhesion lower than 100 kPa, whereas others claim it when the accretion of ice is significantly prolonged. Herein we propose to use a terminology based on the type of frozen water deposition and its physical behaviour, separating ice and snow in adhesion and accretion. This is done in order to lay a foundation on which to introduce a framework for classification of snow- and icephobic surfaces. Such a classification empowers users and producers of such surfaces to communicate accurately about performances and expectations of surfaces and coatings. We include snowphobicity in this scheme, as snow is closely related to the traditional icephobicity and often exhibit overlapping requirements.

Dilated multi-scale cascade forest for satellite image classification

ABSTRACT Satellite image classification is the core of satellite image interpretation. The traditional shallow learning classification algorithm is difficult to extract remote sensing image features effectively, resulting in low classification accuracy. The deep neural network can automatically realize the feature analysis. However, due to the similarity of the optical properties of the cloud and snow, the existing deep learning model is difficult to obtain satisfactory results for the cloud and snow classification in the plateau. Meanwhile, too many parameters of deep learning lead to slow detection speed, which is difficult to meet the real-time requirements of detection. Furthermore, the deep learning method is prone to model degradation. In view of the existing problems, this paper proposes a dilated multi-scale cascade forest method to realize the classification of satellite cloud images. Multi-scale scanning increases the diversity of feature vector extraction and improves the separability of features. The dilated structure is designed to expand the feature receptive field, it improves the efficiency of feature extraction without losing the texture and spatial correlation information of the original image. In addition, the number of layers of network cascading does not need to be customized, it can be determined automatically according to the performance of the system. The proposed method effectively improves the accuracy of cloud/snow classification. Experiments on the imagery of HuanJing satellites A and B (HJ-1A/1B) in China demonstrate that the proposed model can get better classification results than existing studies. The time complexity of the proposed model is lower than existing works, hence it is practical in real-time and accurate remote sensing image classification.