Multispectral Datasets Research Articles

Estimating Carbon Stock in Unmanaged Forests Using Field Data and Remote Sensing

Unmanaged forest ecosystems play a critical role in addressing the ongoing climate and biodiversity crises. As there is no commercial interest in monitoring the health and development of such inaccessible habitats, low-cost assessment approaches are needed. We used a method combining RGB imagery acquired using an Unmanned Aerial Vehicle (UAV), Sentinel-2 data, and field surveys to determine the carbon stock of an unmanaged forest in the UNESCO World Heritage Site wilderness area Dürrenstein-Lassingtal in Austria. The entry-level consumer drone (DJI Mavic Mini) and freely available Sentinel-2 multispectral datasets were used for the evaluation. We merged the Sentinel-2 derived vegetation index NDVI with aerial photogrammetry data and used an orthomosaic and a Digital Surface Model (DSM) to map the extent of woodland in the study area. The Random Forest (RF) machine learning (ML) algorithm was used to classify land cover. Based on the acquired field data, the average carbon stock per hectare of forest was determined to be 371.423 ± 51.106 t of CO2 and applied to the ML-generated class Forest. An overall accuracy of 80.8% with a Cohen’s kappa value of 0.74 was achieved for the land cover classification, while the carbon stock of the living above-ground biomass (AGB) was estimated with an accuracy within 5.9% of field measurements. The proposed approach demonstrated that the combination of low-cost remote sensing data and field work can predict above-ground biomass with high accuracy. The results and the estimation error distribution highlight the importance of accurate field data.

LFIR-YOLO: Lightweight Model for Infrared Vehicle and Pedestrian Detection

The complexity of urban road scenes at night and the inadequacy of visible light imaging in such conditions pose significant challenges. To address the issues of insufficient color information, texture detail, and low spatial resolution in infrared imagery, we propose an enhanced infrared detection model called LFIR-YOLO, which is built upon the YOLOv8 architecture. The primary goal is to improve the accuracy of infrared target detection in nighttime traffic scenarios while meeting practical deployment requirements. First, to address challenges such as limited contrast and occlusion noise in infrared images, the C2f module in the high-level backbone network is augmented with a Dilation-wise Residual (DWR) module, incorporating multi-scale infrared contextual information to enhance feature extraction capabilities. Secondly, at the neck of the network, a Content-guided Attention (CGA) mechanism is applied to fuse features and re-modulate both initial and advanced features, catering to the low signal-to-noise ratio and sparse detail features characteristic of infrared images. Third, a shared convolution strategy is employed in the detection head, replacing the decoupled head strategy and utilizing shared Detail Enhancement Convolution (DEConv) and Group Norm (GN) operations to achieve lightweight yet precise improvements. Finally, loss functions, PIoU v2 and Adaptive Threshold Focal Loss (ATFL), are integrated into the model to better decouple infrared targets from the background and to enhance convergence speed. The experimental results on the FLIR and multispectral datasets show that the proposed LFIR-YOLO model achieves an improvement in detection accuracy of 4.3% and 2.6%, respectively, compared to the YOLOv8 model. Furthermore, the model demonstrates a reduction in parameters and computational complexity by 15.5% and 34%, respectively, enhancing its suitability for real-time deployment on resource-constrained edge devices.

Classifying early apple scab infections in multispectral imagery using convolutional neural networks

Multispectral imaging systems combined with deep learning classification models can be cost-effective tools for the early detection of apple scab (Venturia inaequalis) disease in commercial orchards. Near-infrared (NIR) imagery can display apple scab symptoms earlier and at a greater severity than visible-spectrum (RGB) imagery. Early apple scab diagnosis based on NIR imagery may be automated using deep learning convolutional neural networks (CNNs). CNN models have previously been used to classify a range of apple diseases accurately but have primarily focused on identifying late-stage rather than early-stage detection. This study fine-tunes CNN models to classify apple scab symptoms as they progress from the early to late stages of infection using a novel multispectral (RGB-NIR) time series created especially for this purpose.This novel multispectral dataset was used in conjunction with a large Apple Disease Identification (ADID) dataset created from publicly available, pre-existing disease datasets. This ADID dataset contained 29,000 images of infection symptoms across six disease classes. Two CNN models, the lightweight MobileNetV2 and heavyweight EfficientNetV2L, were fine-tuned and used to classify each disease class in a testing dataset, with performance assessed through metrics derived from confusion matrices. The models achieved scab-prediction accuracies of 97.13 % and 97.57 % for MobileNetV2 and EfficientNetV2L, respectively, on the secondary data but only achieved accuracies of 74.12 % and 78.91 % when applied to the multispectral dataset in isolation. These lower performance scores were attributed to a higher proportion of false-positive scab predictions in the multispectral dataset. Time series analyses revealed that both models could classify apple scab infections earlier than the manual classification techniques, leading to more false-positive assessments, and could accurately distinguish between healthy and infected samples up to 7 days post-inoculation in NIR imagery.

Assessment of the spaceborne EnMAP hyperspectral data for alteration mineral mapping: A case study of the Reko Diq porphyry Cu[sbnd]Au deposit, Pakistan

For over four decades, spaceborne multispectral data have played a crucial role in supporting mineral exploration and geologic mapping. The spaceborne multispectral datasets, however, have a restricted number of bands with coarse spectral resolution and, thus are very limited in mineral mapping. The advent of high-quality spaceborne imaging spectroscopic data like the Environmental Mapping and Analysis Program (EnMAP), has bridged this gap initiating a new era in global hyperspectral mineral mapping. The EnMAP satellite, operational since November 2022, covers the spectral range of 420 and 2450 nm in 224 bands, offering a spatial resolution of 30 m and a mean spectral sampling distance of 8.1 and 12.5 nm in the visible-near infrared and shortwave infrared regions, respectively. In this paper, we demonstrate the enhanced mapping capabilities of EnMAP using datasets acquired over the Reko Diq mining district, a cluster of Miocene porphyries located in Pakistan's Chagai Belt hosting an undeveloped world-class porphyry Cu-Au ± Mo deposit. The EnMAP's Level 2A data product was processed using the polynomial fitting technique to characterize the diagnostic absorption features of the alteration minerals in the Reko Diq porphyry system. This involved retrieving the minimum wavelength, depth, width, and asymmetry parameters for key absorption features and employing them interactively for mineral characterization. A diverse array of minerals were successfully mapped over the study area and validated by ground spectroscopy. This includes abundance/composition maps for white micas, chlorite, epidote, calcite, kaolinite, gypsum, jarosite, and ferric and ferrous iron minerals. The minimum wavelength of white mica was found to vary between 2195 and 2210 nm, with the shorter wavelengths (Al-rich) white mica occurring proximal to the known mineralized zones. The potassic alteration cores and the outer propylitic zones were identified by the ferrous iron and chlorite-epidote-calcite mineral maps, respectively. This study demonstrated the superiority of EnMAP hyperspectral data in delineating the alteration mineralogy and zonation pattern of porphyry copper systems. This capability can potentially contribute to the exploration of new deposits in exposed terrains worldwide.

Analysis of a large buried impact crater and vertical mineral composition at the Chang'E-4 landing site by multi-source remote sensing data

Exploring the concealed subsurface structures and materials beneath the lunar surface can reveal significant insights into geological history. This study offers a comprehensive analysis of the stratigraphic interpretation and subsurface material composition at the Chang'E-4 landing site, integrating both in-situ and orbital radar with multispectral datasets. We report the identification of a subsurface structure, which resembles a buried impact crater (∼420 m in diameter) under the Yutu-2 rover's path. This crater could degrade over a period of 0.42 to 0.53 Ga, with an initial diameter of 293 to 323 m and an initial depth of 45.9 to 51.4 m. Surface material above the buried crater, evaluated by the in-situ visible and near-infrared imaging spectrometer (VNIS) detector, shows a higher abundance of clinopyroxene compared to surrounding areas, where a near-equal mix of clinopyroxene and orthopyroxene is observed. Assessment of crater diameters in proximity to the Chang'E-4 landing site, along with the mineral compositions at their epicenters, reveals a decrease in the abundance of clinopyroxene and plagioclase with depth. Conversely, the quantities of orthopyroxene and olivine increase, implying that orthopyroxene-rich Finsen ejecta significantly influenced the Chang'E-4 landing site's geological composition. Two potential stratigraphic boundary depths are identified at 13.5 and 22 m, based on pronounced variations in mineral abundance, offering fresh insights into subsurface delineation beyond radar data. Considering the VNIS and vertical mineral composition, we propose the buried crater's formation resulted from Finsen crater's ejecta. Also, we identify eight potential historical impacts by comparing subsurface relief variations with mineral composition ratios between clinopyroxene and orthopyroxene. The integration of subsurface structure, along with surface and subsurface mineral composition, enables a more robust stratigraphic interpretation, facilitates shallow material source analysis, and allows for historical impact tracing.

دراسة مقارنة التعلم الآلي في التصنيف الفعال للبيانات متعددة الأطياف في الزراعة

يتطلب أن تكون خرائط محاصيل موثوقة ودقيقة لتحقيق الأمن الغذائي من المستوى الإقليمي إلى المستوى العالمي. يؤدي التوفر المتزايد لصور الأقمار الصناعية إلى مشكلة "البيانات الضخمة" أثناء إنتاج خرائط المحاصيل. الآن، اكتسبت المنصات السحابية الكثير من الاهتمام لتصنيف المحاصيل في مناطق واسعة. الهدف الرئيسي من البحث هو تحليل تصنيف المحاصيل باستخدام مختلف التعلم الآلي (ML) مثل آلة دعم المتجهات (SVM)، وتعزيز شجرة التدرج (GTB)، والغابات العشوائية (RF)، وشجرة القرار (DT) بالإضافة إلى التصنيف والتصنيف. أشجار الانحدار (CART) على منصة محرك Goggle Earth. الهدف هو استكشاف كفاءة محرك Google Earth (GEE) عند تصنيف المحاصيل المختلفة باستخدام مجموعات البيانات متعددة الأطياف من Sentinel 2 MSI بالإضافة إلى الأقمار الصناعية Landsat 8 OLI لرسم خرائط المحاصيل في منطقة ماثورا في ولاية أوتار براديش بالهند. تم استخدام أفضل صورة خالية من السحابة (أقل من 5%) لمجموعات بيانات Landsat 8 OLI وSentinel 2 MSI ("2020-12-26"، و"2020-12-30") لتصنيف المحاصيل بمساعدة التصفية التلقائية، أي النسبة المئوية. الملكية السحابية على منصات GEE. علاوة على ذلك، يمكن تنظيم أداء منصة GEE والحصول عليها وتوضيحها وكذلك معالجتها المسبقة لمجموعة بيانات الأقمار الصناعية بقوة كبيرة. تم استخدام النقاط كمساحات مميزة مثل مجموعات بيانات التدريب. علاوة على ذلك، يتم استخدام مصفوفات الارتباك لتقييم الدقة (دقة المنتج والمستخدم) ومعامل كابا. بالإضافة إلى ذلك، قم بمقارنة نتائج مجموعة البيانات (Sentinel 2 MSI وLandsat 8 OLI) على أساس الدقة الإجمالية (OA) ودرجة F1 بالإضافة إلى معامل كابا. تم العثور على أعلى وصول حر باستخدام GTB (86.7%) يليه RF (82.5%)، CART (81.0%)، DT (78.1%) وSVM (66.5%) لصورة Landsat 8 OLI. بالنسبة لصورة Sentinel 2، حققت GTB أعلى وصول وصول بنسبة 84.2% يليها SVM (84%)، RF (82.3%)، DT (75.2%)، وCART (75.0%) على التوالي. على أساس البحث، وجد أن أداء GTB كان جيدًا بين جميع المصنفات في رسم خرائط المحاصيل باستخدام مجموعتي البيانات متعددة الأطياف.

Developing a Semi-Automated Near-Coastal, Water Quality-Retrieval Process from Global Multi-Spectral Data: South-Eastern Australia

The estimation of water quality properties through satellite remote sensing relies on (1) the optical characteristics of the water body, (2) the resolutions (spatial, spectral, radiometric and temporal) of the sensor and (3) algorithm(s) applied. More than 80% of global water bodies fall under Case I (open ocean) waters, dominated by scattering and absorption associated with phytoplankton in the water column. Globally, previous studies show significant correlations between satellite-based retrieval methods and field measurements of absorbing and scattering constituents, while limited research from Australian coastal water bodies appears. This study presents a methodology to extract chlorophyll a properties from surface waters from near-coastal environments, within 2 km of coastline, in Tasmania, south-eastern Australia. We use general purpose, global, long-time series, multi-spectral satellite data, as opposed to ocean colour-specific sensor data. This approach may offer globally applicable tools for combining global satellite image archives with in situ field sensors for water quality monitoring. To enable applications from local to global scales, a cloud-based geospatial analysis workflow was developed and tested on several sites. This work represents the initial stage in developing a semi-automated near-coastal water-quality workflow using easily accessed, fully corrected global multi-spectral datasets alongside large-scale computation and delivery capabilities. Our results indicated a strong correlation between the in situ chlorophyll concentration data and blue-green band ratios from the multi-spectral sensor. In line with published research, environment-specific empirical models exhibited the highest correlations between in situ and satellite measurements, underscoring the importance of tailoring models to specific coastal waters. Our findings may provide the basis for developing this workflow for other sites in Australia. We acknowledge the use of general purpose multi-spectral data such as the Sentinel-2 and Landsat Series, their corrections and algorithms may not be as accurate and precise as ocean colour satellites. The data we are using are more readily accessible and also have true global coverage with global historic archives and regular, global collection will continue at least 10 years in the future. Regardless of sensor specifications, the retrieval method relies on localised algorithm calibration and validation using in situ measurements, which demonstrates close-to-realistic outputs. We hope this approach enables future applications to also consider these globally accessible and regularly updated datasets that are suited to coastal environments.

Comparative assessment of land use land cover at different temporal scales in Imphal-Iril river catchment, Manipur, India

ABSTRACT Land use land cover (LULC) assessment is essential for monitoring naturally available land resources through spatio-temporal changing pattern to protect land resources sustainably. The primary objective is to identify the changing pattern in LULC over last 10 years in Imphal-Iril catchment. Multi-spectral datasets of Landsat 7 and 8 were used for five different time periods (2011, 2013, 2015, 2017 and 2019). Supervised classification method was employed to generate the classified LULC maps using the maximum likelihood classifier approach in ArcGIS 10.3. In this assessment, five major land use classes have been classified as vegetation, agriculture, settlement, barren land and water body, and the dominant land cover is vegetation. The classified images obtained from this technique were checked for their accuracy using statistical kappa co-efficient and overall accuracy. According to this study, settlement and barren areas have been significantly increased by 72.25% (39.03 km2) and 22.60% (118.29 km2) during the last 10 years, whereas areas assigned to agriculture, vegetation and water bodies have decreased by 27.26% (72.16 km2), 7.59% (73.22 km2) and 60.39% (11.94 km2), respectively. LULC map for the year 2030 was also predicted using MOLUSCE model in QGIS. The results of this assessment will be significant to the policy implicates for sustainable LULC management in the region.

Time division multiplexing based multi-spectral semantic camera for LiDAR applications

The recent progress in the development of measurement systems for autonomous recognition had a substantial impact on emerging technology in numerous fields, especially robotics and automotive applications. In particular, time-of-flight (TOF) based light detection and ranging (LiDAR) systems enable to map the surrounding environmental information over long distances and with high accuracy. The combination of advanced LiDAR with an artificial intelligence platform allows enhanced object recognition and classification, which however still suffers from limitations of inaccuracy and misidentification. Recently, multi-spectral LiDAR systems have been employed to increase the object recognition performance by additionally providing material information in the short-wave infrared (SWIR) range where the reflection spectrum characteristics are typically very sensitive to material properties. However, previous multi-spectral LiDAR systems utilized band-pass filters or complex dispersive optical systems and even required multiple photodetectors, adding complexity and cost. In this work, we propose a time-division-multiplexing (TDM) based multi-spectral LiDAR system for semantic object inference by the simultaneous acquisition of spatial and spectral information. By utilizing the TDM method, we enable the simultaneous acquisition of spatial and spectral information as well as a TOF based distance map with minimized optical loss using only a single photodetector. Our LiDAR system utilizes nanosecond pulses of five different wavelengths in the SWIR range to acquire sufficient material information in addition to 3D spatial information. To demonstrate the recognition performance, we map the multi-spectral image from a human hand, a mannequin hand, a fabric gloved hand, a nitrile gloved hand, and a printed human hand onto an RGB-color encoded image, which clearly visualizes spectral differences as RGB color depending on the material while having a similar shape. Additionally, the classification performance of the multi-spectral image is demonstrated with a convolution neural network (CNN) model using the full multi-spectral data set. Our work presents a compact novel spectroscopic LiDAR system, which provides increased recognition performance and thus a great potential to improve safety and reliability in autonomous driving.

Using Unmanned Aerial Vehicles and Multispectral Sensors to Model Forage Yield for Grasses of Semiarid Landscapes

Forage yield estimates provide relevant information to manage and quantify ecosystem services in grasslands. We fitted and validated prediction models of forage yield for several prominent grasses used in restoration projects in semiarid areas. We used field forage harvests from three different sites in Northern Utah and Southern California, USA, in conjunction with multispectral, high-resolution UAV imagery. Different model structures were tested with simple models using a unique predictor, the forage volumetric 3D space, and more complex models, where RGB, red edge, and near-infrared spectral bands and associated vegetation indices were used as predictors. We found that for most dense canopy grasses, using a simple linear model structure could explain most (R2 0.7) of the variability of the response variable. This was not the case for sparse canopy grasses, where a full multispectral dataset and a non-parametric model approach (random forest) were required to obtain a maximum R2 of 0.53. We developed transparent protocols to model forage yield where, in most circumstances, acceptable results could be obtained with affordable RGB sensors and UAV platforms. This is important as users can obtain rapid estimates with inexpensive sensors for most of the grasses included in this study.

Scanning Micro X-ray Fluorescence and Multispectral Imaging Fusion: A Case Study on Postage Stamps.

Scanning micrο X-ray fluorescence (μ-XRF) and multispectral imaging (MSI) were applied to study philately stamps, selected for their small size and intricate structures. The μ-XRF measurements were accomplished using the M6 Jetstream Bruker scanner under optimized conditions for spatial resolution, while the MSI measurements were performed employing the XpeCAM-X02 camera. The datasets were acquired asynchronously. Elemental distribution maps can be extracted from the μ-XRF dataset, while chemical distribution maps can be obtained from the analysis of the multispectral dataset. The objective of the present work is the fusion of the datasets from the two spectral imaging modalities. An algorithmic co-registration of the two datasets is applied as a first step, aiming to align the multispectral and μ-XRF images and to adapt to the pixel sizes, as small as a few tens of micrometers. The dataset fusion is accomplished by applying k-means clustering of the multispectral dataset, attributing a representative spectrum to each pixel, and defining the multispectral clusters. Subsequently, the μ-XRF dataset within a specific multispectral cluster is analyzed by evaluating the mean XRF spectrum and performing k-means sub-clustering of the μ-XRF dataset, allowing the differentiation of areas with variable elemental composition within the multispectral cluster. The data fusion approach proves its validity and strength in the context of philately stamps. We demonstrate that the fusion of two spectral imaging modalities enhances their analytical capabilities significantly. The spectral analysis of pixels within clusters can provide more information than analyzing the same pixels as part of the entire dataset.

Prediction of species richness and diversity in sub‐alpine grasslands using satellite remote sensing and random forest machine‐learning algorithm

AbstractAimsRemote‐sensing approaches could be beneficial for monitoring and compiling essential biodiversity data because they are cost‐effective and allow for coverage of large areas over a short period. This study investigated the relationship between multispectral remote‐sensing data from Landsat 8 and Sentinel‐2 and species richness and diversity in mountainous and protected grasslands.LocationsGolden Gate Highlands National Park, Free State, South Africa.MethodsIn‐situ data of plant species composition and cover from 142 plots with 16 relevés each were distributed across the study site and used to calculate species richness and Shannon–Wiener species diversity index (species diversity). We used a machine‐learning random forest algorithm to optimize the prediction of species richness and diversity. The algorithm was used to identify the optimal spectral bands and vegetation indices for estimating species richness and diversity. Subsequently, the selected bands and vegetation indices were used to estimate species richness through random forest regression.ResultsThis research found weak relationships between remote‐sensing vegetation indices and the diversity metrics, but significant relationships were found between some spectral bands and diversity metrics. Moreover, using machine‐learning random forest, the multispectral data sets exhibited strong predictive powers. In this investigation, near‐infrared (NIR) seemed to be the most selected band for both sensors to explain species diversity in mountainous grasslands.Main conclusionsThis finding further ascertains the efficiency of optimizing high spatial resolution spectral information to estimate plant species richness and diversity. This research shows that NIR, Soil‐Adjusted Vegetation Index (SAVI) and Enhanced Vegetation Index (EVI) are the most adequate for predicting species richness and diversity in mountainous grasslands with relatively good accuracies. Plant phenology and the choice of sensor affect the relationship between spectral information and species diversity variables.

An applied framework to unlocking multi-angular UAV reflectance data: a case study for classification of plant parameters in maize (Zea mays)

AbstractOptical sensors, mounted on uncrewed aerial vehicles (UAVs), are typically pointed straight downward to simplify structure-from-motion and image processing. High horizontal and vertical image overlap during UAV missions effectively leads to each object being measured from a range of different view angles, resulting in a rich multi-angular reflectance dataset. We propose a method to extract reflectance data, and their associated distinct view zenith angles (VZA) and view azimuth angles (VAA), from UAV-mounted optical cameras; enhancing plant parameter classification compared to standard orthomosaic reflectance retrieval. A standard (nadir) and a multi-angular, 10-band multispectral dataset was collected for maize using a UAV on two different days. Reflectance data was grouped by VZA and VAA (on average 2594 spectra/plot/day for the multi-angular data and 890 spectra/plot/day for nadir flights only, 13 spectra/plot/day for a standard orthomosaic), serving as predictor variables for leaf chlorophyll content (LCC), leaf area index (LAI), green leaf area index (GLAI), and nitrogen balanced index (NBI) classification. Results consistently showed higher accuracy using grouped VZA/VAA reflectance compared to the standard orthomosaic data. Pooling all reflectance values across viewing directions did not yield satisfactory results. Performing multiple flights to obtain a multi-angular dataset did not improve performance over a multi-angular dataset obtained from a single nadir flight, highlighting its sufficiency. Our openly shared code (https://github.com/ReneHeim/proj_on_uav) facilitates access to reflectance data from pre-defined VZA/VAA groups, benefiting cross-disciplinary and agriculture scientists in harnessing the potential of multi-angular datasets. Graphical abstract

A One-Class Classifier for the Detection of GAN Manipulated Multi-Spectral Satellite Images

The current image generative models have achieved a remarkably realistic image quality, offering numerous academic and industrial applications. However, to ensure these models are used for benign purposes, it is essential to develop tools that definitively detect whether an image has been synthetically generated. Consequently, several detectors with excellent performance in computer vision applications have been developed. However, these detectors cannot be directly applied as they areto multi-spectral satellite images, necessitating the training of new models. While two-class classifiers generally achieve high detection accuracies, they struggle to generalize to image domains and generative architectures different from those encountered during training. In this paper, we propose a one-class classifier based on Vector Quantized Variational Autoencoder 2 (VQ-VAE 2) features to overcome the limitations of two-class classifiers. We start by highlighting the generalization problem faced by binary classifiers. This was demonstrated by training and testing an EfficientNet-B4 architecture on multiple multi-spectral datasets. We then illustrate that the VQ-VAE 2-based classifier, which was trained exclusively on pristine images, could detect images from different domains and generated by architectures not encountered during training. Finally, we conducted a head-to-head comparison between the two classifiers on the same generated datasets, emphasizing the superior generalization capabilities of the VQ-VAE 2-based detector, wherewe obtained a probability of detection at a 0.05 false alarm rate of 1 for the blue and red channels when using the VQ-VAE 2-based detector, and 0.72 when we used the EfficientNet-B4 classifier.

A big data driven vegetation disease and pest region identification method based on self supervised convolutional neural networks and parallel extreme learning machines

A self supervised convolutional neural network-parallel extreme learning machine classification model based on big data is proposed to address the subjectivity and inaccuracy of traditional methods for identifying vegetation pests and diseases that rely on manual observation and empirical judgment. This model is constructed using convolutional neural networks and parallel extreme learning machines, and integrates feature extraction networks with dual attention mechanisms to improve the accuracy of identifying pests and diseases. The model utilized a large amount of big data for training, achieving a recall rate of 98.42 % on multispectral datasets, and an overall classification accuracy of 99.04 %. After optimizing the residual network, the overall accuracy of identifying vegetation pest and disease areas has been further improved to 99.77 %, and the recall rate has also reached 98.91 %. These results indicate that the method proposed in this study has high accuracy and efficiency in the application of big data, can meet the needs of disease and pest identification, and provides effective technical support for the monitoring and prevention of crop diseases and pests, which has important practical significance.

INSANet: INtra-INter Spectral Attention Network for Effective Feature Fusion of Multispectral Pedestrian Detection.

Pedestrian detection is a critical task for safety-critical systems, but detecting pedestrians is challenging in low-light and adverse weather conditions. Thermal images can be used to improve robustness by providing complementary information to RGB images. Previous studies have shown that multi-modal feature fusion using convolution operation can be effective, but such methods rely solely on local feature correlations, which can degrade the performance capabilities. To address this issue, we propose an attention-based novel fusion network, referred to as INSANet (INtra-INter Spectral Attention Network), that captures global intra- and inter-information. It consists of intra- and inter-spectral attention blocks that allow the model to learn mutual spectral relationships. Additionally, we identified an imbalance in the multispectral dataset caused by several factors and designed an augmentation strategy that mitigates concentrated distributions and enables the model to learn the diverse locations of pedestrians. Extensive experiments demonstrate the effectiveness of the proposed methods, which achieve state-of-the-art performance on the KAIST dataset and LLVIP dataset. Finally, we conduct a regional performance evaluation to demonstrate the effectiveness of our proposed network in various regions.

Identification of vulnerable areas using geospatial technologies in the lower Manair River basin of Telangana, Southern India

The present study is undertaken at the lower Manair River basin, the central part of Telangana in southern India, as the Government authorities has done rapid developmental activities over the last seven years demanded identifying change detection in environmental indicators. Geospatial tools, viz., remote sensing and geographic information system, are used to classify and identify land-use/land cover (LULC) changes during 2015–2021 based on the images collected from Landsat 8 OLI/TIRS multispectral and multi-temporal satellite data set. Change detection of environmental indicators such as LULC changes, land surface temperature (LST), and normalized difference vegetation index (NDVI) are crucial parameters that play a vital role in identifying vulnerable areas, which helps in planning and utilizing natural resources. The results revealed that the agricultural cropland filled in the top class with an area of 1124.70 km2 (33.82%) in 2015, amplified to 1442.95 km2 (43.39%) in 2021. The built-up area has marginally raised by 581.45 km2 (17.49%) in the study region. A negative correlation is observed between LST and NDVI. LULC patterns are changing, and its imprint echoed on a surge of LST. It establishes a foundation for evidence-based policymaking and sustainable growth, emphasizing the importance of proactive measures in preserving the delicate balance between development and environmental conservation.

Identifying the optimal waveband positions for autumn grassland senescence detection using the broadband multispectral remotely sensed dataset

While remote sensing of grass senescence is addressed in the literature, knowledge of optimal waveband positions that are suitable for discriminating between senescent and non-senescent grasses is still limited. Notably, detection of senescent grass is important for understanding the available forage in rangeland environments and associated ecological implications. The free provision of remote sensing data from modern broadband multispectral sensors with improved spatial and spectral properties offers prospects for reliable and wall-to-wall monitoring of grassland senescence in rangeland ecosystems. The current study tested the potential of the modern multispectral remote sensing dataset (i.e., Sentinel 2 and Landsat) in mapping the senescent grass, and to identify the optimal waveband positions that are suitable for discriminating between senescent and non-senescent grasses. Locational information for both senescent and non-senescent grasses was acquired on the field and was used to train the classification process. A Random Forest classification approach was employed using the Landsat 8 and Sentinel 2 multispectral datasets to spectrally discern between senescent and non-senescent grasses. Our analysis yielded overall classification accuracies of 0.82 and 0.78 and kappa coefficients of 0.64 and 0.56 for Sentinel 2 and Landsat 8, respectively. Using the stepwise selection approach, the study further identified that the Red Edge Position (REP), and the visible green and red bands of the electromagnetic spectrum were the optimal waveband positions for separating between senescent and non-senescent grasses based on the broadband multispectral remote sensing. This study has demonstrated the value of the broadband multispectral remote sensing data in mapping autumn grassland senescence, and this lays a foundation for effective operational scale monitoring of foraging resources at the landscape scale, particularly during dry periods.

Semi-supervised learning with constrained virtual support vector machines for classification of remote sensing image data

We introduce two semi-supervised models for the classification of remote sensing image data. The models are built upon the framework of Virtual Support Vector Machines (VSVM). Generally, VSVM follow a two-step learning procedure: A Support Vector Machines (SVM) model is learned to determine and extract labeled samples that constitute the decision boundary with the maximum margin between thematic classes, i.e., the Support Vectors (SVs). The SVs govern the creation of so-called virtual samples. This is done by modifying, i.e., perturbing, the image features to which a decision boundary needs to be invariant. Subsequently, the classification model is learned for a second time by using the newly created virtual samples in addition to the SVs to eventually find a new optimal decision boundary. Here, we extend this concept by (i) integrating a constrained set of semi-labeled samples when establishing the final model. Thereby, the model constrainment, i.e., the selection mechanism for including solely informative semi-labeled samples, is built upon a self-learning procedure composed of two active learning heuristics. Additionally, (ii) we consecutively deploy semi-labeled samples for the creation of semi-labeled virtual samples by modifying the image features of semi-labeled samples that have become semi-labeled SVs after an initial model run. We present experimental results from classifying two multispectral data sets with a sub-meter geometric resolution. The proposed semi-supervised VSVM models exhibit the most favorable performance compared to related SVM and VSVM-based approaches, as well as (semi-)supervised CNNs, in situations with a very limited amount of available prior knowledge, i.e., labeled samples.

A Generic Self-Supervised Learning (SSL) Framework for Representation Learning from Spectral–Spatial Features of Unlabeled Remote Sensing Imagery

Remote sensing data has been widely used for various Earth Observation (EO) missions such as land use and cover classification, weather forecasting, agricultural management, and environmental monitoring. Most existing remote-sensing-data-based models are based on supervised learning that requires large and representative human-labeled data for model training, which is costly and time-consuming. The recent introduction of self-supervised learning (SSL) enables models to learn a representation from orders of magnitude more unlabeled data. The success of SSL is heavily dependent on a pre-designed pretext task, which introduces an inductive bias into the model from a large amount of unlabeled data. Since remote sensing imagery has rich spectral information beyond the standard RGB color space, it may not be straightforward to extend to the multi/hyperspectral domain the pretext tasks established in computer vision based on RGB images. To address this challenge, this work proposed a generic self-supervised learning framework based on remote sensing data at both the object and pixel levels. The method contains two novel pretext tasks, one for object-based and one for pixel-based remote sensing data analysis methods. One pretext task is used to reconstruct the spectral profile from the masked data, which can be used to extract a representation of pixel information and improve the performance of downstream tasks associated with pixel-based analysis. The second pretext task is used to identify objects from multiple views of the same object in multispectral data, which can be used to extract a representation and improve the performance of downstream tasks associated with object-based analysis. The results of two typical downstream task evaluation exercises (a multilabel land cover classification task on Sentinel-2 multispectral datasets and a ground soil parameter retrieval task on hyperspectral datasets) demonstrate that the proposed SSL method learns a target representation that covers both spatial and spectral information from massive unlabeled data. A comparison with currently available SSL methods shows that the proposed method, which emphasizes both spectral and spatial features, outperforms existing SSL methods on multi- and hyperspectral remote sensing datasets. We believe that this approach has the potential to be effective in a wider range of remote sensing applications and we will explore its utility in more remote sensing applications in the future.