PSWGAN-GP:改进梯度惩罚的生成对抗网络

生成对抗网络的出现对解决深度学习领域样本数据不足的研究起到了极大的促进作用。为解决生成对抗网络生成的图像出现轮廓模糊、前景背景分离等细节质量问题,提出一种改进梯度惩罚的Wasserstein生成对抗网络算法(PSWGAN-GP)。该算法在WGAN-GP的Wasserstein距离损失和梯度惩罚的基础上,在判别器中使用从VGG-16网络的3个池化层中提取的特征,并通过这些特征计算得出风格损失(Style-loss)和感知损失(Perceptual-loss)作为原损失的惩罚项,提升判别器对深层特征的获取和判别能力,对生成图像的细节进行修正和提升。实验结果表明,在生成器和判别器网络结构相同,并保证超参数相同的情况下,PSWGAN-GP的IS评分和FID评分相对于参与对比的其他图像生成算法有所提升,且可有效改善生成图片的细节质量。     

Measuring fire severity using UAV imagery in semi-arid central Queensland,Australia

Remote-sensing methods for fire severity mapping have traditionally relied on multispectral imagery captured by satellite platforms carrying passive sensors such as Landsat Thematic Mapper /Enhanced Thematic Mapper Plus or Moderate Resolution Imaging Spectroradiometer. This article describes the analysis of high spatial resolution Unmanned Aerial Vehicle (UAV) imagery to assess fire severity on a 117 ha experimental fire conducted on coal mine rehabilitation in an open woodland environment in semi-arid Central Queensland, Australia. Three band indices, Excess Green Index, Excess Green Index Ratio, and Modified Excess Green Index, were used to derive differenced (d) fire severity maps from UAV data. Fire severity data sets derived from aerial photograph interpretation were used to assess the utility of employing UAV technology to determine fire severity impacts. The dEGI was able to separate high severity, low severity, and unburnt areas with an overall classification accuracy of 58% and Kappa statistic of 0.37; outperforming the dEGIR (overall accuracy 55%, Kappa 0.31) and the dMEGI (overall accuracy 38%, Kappa 0.06). Classification accuracy increased for all indices when canopy shadows were masked, with dEGI improving to an overall accuracy of 68% and 0.48 Kappa. The McNemar’s test indicated that there was no significant difference between the classification accuracies for dEGI and dEGIR (p < 0.05). The test also demonstrated that dMEGI was significantly lower in accuracy compared to dEGI and dEGIR (p < 0.05). We quantified the proportion of burnt area within each severity class and calculated that 32% of the site was burnt at high severity, 34% was burnt at low severity, and 34% of the block was unburnt due to the patchy nature of the fire. We discuss the UAV-specific errors associated with fire severity mapping, and the potential for UAVs to assist land managers to assess the extent and severity of fire and subsequent recovery of burnt ecosystems at local scales (10⁴m²–1 km²).     

遥感图像森林林型SVM分类的多特征选择

为了研究遥感图像森林林型SVM分类多特征的选择对提高分类精度的影响,选取小波变换不同尺度纹理、四种植被指数、最优波段光谱特征等不同组合构成林型分类多特征向量进行分类。结果表明,纹理与植被指数、最优波段组合多特征的森林林型分类精度最高,阔叶林、针叶林和竹林的分类精度分别为84.4%、86.5%、91.0%,比纹理单类特征分类分别提高4.1%、4.0%、1.1%,比植被指数单类特征分类分别提高9.2%、11.8%、11.9%。多特征的分类精度一般要高于单类特征,纹理能够较明显提高林型可分性,植被指数也有一定的效果,但最优波段光谱特征的效果较弱。     

Ultra‐wide band antenna attachable on UAV surface

In the present study, ultra‐wide band antenna attachable on unmanned aerial vehicle (UAV) surfaces usable as signal detection antenna in various bands was designed, and the practicality of the developed antenna was verified by attaching the antenna to a UAV and measuring the performance. The antenna suggested in this article was manufactured by forming a hemispheric conductor having the shape of a baseball seam on the ground and satisfying a self‐complementary through an image theory. This Hemispheric shape can reduce a drag and risk of breakage. The diameter of the antenna is 400 mm (0.4λ_L and λ_L is 0.3 GHz.), the height is 200 mm (0.2λ_L), and the ground size is 800 mm (0.8λ_L) × 800 mm (0.8λ_L). The designed antenna showed an ultra‐wide band property as it was matched to a band from 300 MHz to 10 GHz. After being attached to the bottom of a 7.3:1 scale UAV, the antenna showed matching properties from 1.85 to 10 GHz above and maintained a monopole pattern in all directions and in a bandwidth. To author's knowledge, the antenna was proper to using as a signal detection antenna which need wide bandwidth, conical pattern, drag reduction and reduced risk of breakage.     

Assessing leaf scale measurement for nitrogen content of oil palm: performance of discriminant analysis and Support Vector Machine classifiers

Nitrogen (N) is a crucial element in sustaining oil palm production. However, assessing N status of tall perennial crops such as oil palm is complex and not as straightforward as assessing annual crops, due to complex N partitioning, age, and larger amounts of respiratory loads. Hence, the objectives of this study were to evaluate the potential of spectral measurements obtained from leaf scale and machine learning approaches as a rapid tool for quantifying oil palm N status. This study involved assessing the performance of discriminant analysis (DA) and Support Vector Machine (SVM) classifiers for discriminating spectral bands sensitive to N sufficiency levels and comparing the predictive accuracy of DA and SVM for classifying N status of immature and mature oil palms. The experiment was conducted on immature Tenera seedlings (13 months old) and mature Tenera palm stands (9 and 12 years old) that were arranged in Randomized Complete Block Design with treatments varied from 0 to 2 kg N. Generally, the discriminant function of both classifiers was age-dependent. A clear trade-off between the classifiers’ number of spectral bands and their accuracies was observed; the DA with a larger number of optimal spectral bands could discriminate N sufficiency levels of all maturity classes with higher accuracies compared to the SVM, yet the latter could produce reasonable accuracies with a lesser number of spectral bands. N status of all maturity classes could be classified satisfactorily with SVM (71–88%) via the satellite-simulated blue and green bands, signifying the possibility to develop spectral index or an N-sensitive sensor for oil palm.     

Estimating the nitrogen nutrition index in grass seed crops using a UAV-mounted multispectral camera

The application of adequate nitrogen (N) fertilizers to grass seed crops is important to achieve high seed yield. Application of N will inevitably result in over-fertilization on some fields and, concomitantly, an increased risk of adverse environmental impacts, such as ground- and/or surface-water contamination. This study was designed to estimate the N status of two grass seed crops: red fescue (Festuca rubra L.) and perennial ryegrass (Lolium perenne L.) using images captured with an unmanned aerial vehicle (UAV) mounted multispectral camera. Two types of UAV, a fixed-wing UAV and a multi-rotor UAV, operating at two different heights and mounted with the same multispectral camera, were used in different field experiments at the same location in Denmark in the period from 432 to 861 growing degree-days. Seven vegetation indices, calculated from multispectral images with four bands: red, green, red edge and near infrared (NIR), were evaluated for their relationship to dry matter (DM), N concentration, N uptake and N nutrition index (NNI). The results showed a better prediction of N concentration, N uptake and NNI, than DM using vegetation indices. Furthermore, among all vegetation indices, two red-edge-based indices, normalized difference red edge (NDRE) and red edge chlorophyll index (CI_RE), performed best in estimating N concentration (R² = 0.69–0.88), N uptake (R² = 0.41–0.84) and NNI (R² = 0.47–0.86). In addition, there was no effect from the choice of UAV, and thereby flight height, on the estimation of NNI. The choice of UAV type therefore seems not to influence the possibility of diagnosing N status in grass seed crops. We conclude that it is possible to estimate NNI based on multispectral images from drone-mounted cameras, and the method could guide farmers as to whether they should apply additional N to the field. We also conclude that further research should focus on estimating the quantity of N to apply and on further developing the method to include more grass species.     

Predicting C3 and C4 grass nutrient variability using in situ canopy reflectance and partial least squares regression

The use of hyperspectral data to estimate forage nutrient content can be a challenging task, considering the multicollinearity problem, which is often caused by high data dimensionality. We predicted some variability in the concentration of limiting nutrients such as nitrogen (N), crude protein (CP), moisture, and non-digestible fibres that constrain the intake rate of herbivores. In situ hyperspectral reflectance measurements were performed at full canopy cover for C3 and C4 grass species in a montane grassland environment. The recorded spectra were resampled to 13 selected band centres of known absorption and/or reflectance features, WorldView-2 band settings, and to 10 nm-wide bandwidths across the 400–2500 nm optical region. The predictive accuracy of the resultant wavebands was assessed using partial least squares regression (PLSR) and an accompanying variable importance (VIP) projection. The results indicated that prediction accuracies ranging from 66% to 32% of the variance in N, CP, moisture, and fibre concentrations can be achieved using the spectral-only information. The red, red-edge, and shortwave infrared (SWIR) wavelength regions were the most sensitive to all nutrient variables, with higher VIP values. Moreover, the PLSR model constructed based on spectra resampled around the 13 preselected band centres yielded the highest sensitivity to the predicted nutrient variables. The results of this study thus suggest that the use of the spectral resampling technique that uses only a few but strategically selected band centres of known absorption or reflectance features is sufficient for forage nutrient estimation.     

Barley yield and fertilization analysis from UAV imagery: a deep learning approach

In a world with great challenges in food security, optimising cereal production is critical. Cereals are the most important food source for human consumption. The fourth important cereal worldwide after wheat, rice and maize, is Barley, and its production strongly depends on fertilization treatments. The adoption of suitable fertilizer management strategies often results in large economic benefits to producers. However, determining optimal fertilizer doses for a specific barley variety is complex. The collection of data and their analyses can be cost prohibitive for small farmers regarding time and money. This paper introduces an approach to support producers with automatic tools for the analysis of fertilization management of barley. The proposed methodology aims to simultaneously estimate nitrogen fertilization and barley yield, from information derived from aerial RGB images captured by a UAV. Our long term goal is to provide a low-cost and wide-are-coverage solution for the estimation of barley variables that can be leveraged to increase barley yield without increasing costs. A low-cost UAV is used to capture RGB crop field images. Then, a deep convolutional neural network is used for the automated extraction of features from the images. Extracted features are feed into predictive models that estimate the variables of interest. Experimental results reveal that the proposed methodology is able to reach an accuracy above 83% when estimating nitrogen fertilization and a high correlation and low RMSE in the estimation of yield in grams. Experimental results are promising and will pave the way for the development of deep learning methods for barley analysis from aerial imagery that can be accessed by the average farmer.     

基于无人机遥感技术的台风灾害倒伏绿化树木检测

以福州大学为试验区,提出一种基于无人机遥感影像的台风灾害倒伏绿化树木的快速提取方法,为园林部门进行台风灾害损失评估、灾后重建提供参考.首先利用无人机遥感技术获取高于10 cm分辨率的台风过境前后影像,经过处理得到数字正射影像(Digital Orthophoto Map,DOM)和数字表面模型(Digital Surf...     

Parcel-based urban land use classification in megacity using airborne LiDAR,high resolution orthoimagery,and Google Street View

Urban land use information is increasingly important for a variety of purposes. With their increasing coverage and availability, airborne light detection and ranging (LiDAR) data, high resolution orthoimagery (HRO), and Google Street View (GSV) images are showing great potential for accurate land use classification. However, no study mapped land use in megacity using GSV-derived features or the three kinds of data together for land use classification. The main objectives of this study are (1) to test the performance of a parcel-based land use classification method using a Random Forest classifier with LiDAR data, HRO, and GSV images in a megacity, and (2) to explore the use of GSV in separating parcels of mixed residential & commercial buildings from other land use parcels. Two neighboring community districts in Brooklyn, New York, were selected as the study area. Thirteen automatically-derived parcel features, including nine common parcel features and four GSV-derived parcel features, were used in land use classification. The average overall classification accuracy was 77.5%, with producer's accuracies exceeding 92% for single-family housing. Comparing the results of classifications with and without GSV-derived parcel features shows that GSV-derived parcel features on average contribute to the classification accuracy of mixed residential & commercial buildings by 10 percentage points, improving it from 41.3% to 51.4%. In general, the results show that even in a complex megacity, the parcel-based land use classification technique, with parcel features extracted from airborne LiDAR, HRO, and GSV, is able to discriminate among different land use classes, such as single-family house, commercial & industrial building, and open space & park, with acceptable accuracies, and that integrating GSV into classification improves the classification accuracy of some urban land use classes, especially mixed residential & commercial building.     

Evaluating variations of physiology-based hyperspectral features along a soil water gradient in a Eucalyptus grandis plantation

Remote sensing is viewed as a cost-effective alternative to intensive field surveys in assessing site factors that affect growth of Eucalyptus grandis over broad areas. The objective of this study was to assess the utility of hyperspectral remote sensing to discriminate between site qualities in E. grandis plantation in KwaZulu-Natal, South Africa. The relationships between physiology-based hyperspectral indicators and site quality, as defined by total available water (TAW), were assessed for E. grandis plantations through one-way analysis of variance (ANOVA). Canopy reflectance spectra for 68 trees (25 good, 25 medium and 18 poor sites) were collected on clear-sky days using an Analytical Spectral Device (ASD) spectroradiometer (350–2500 nm) from a raised platform. Foliar macronutrient concentrations for N, P, K, S, Ca, Mg and Na and their corresponding spectral features were also evaluated. The spectral signals for leaf water – normalized difference water index (NDWI), water band index (WBI) and moisture stress index (MSI) – exhibited significant differences (p < 0.05) between sites. The magnitudes of these indices showed distinct gradients from the poor to the good sites. Similar results were observed for chlorophyll indices. These results show that differences in site quality based on TAW could be detected via imaging spectroscopy of canopy water or chlorophyll content. Among the macronutrients, only K and Ca exhibited significant differences between sites. However, a Tukey post-hoc test showed differences between the good and medium or medium and poor sites, a trend not consistent with the TAW gradient. The study also revealed the capability of continuum-removed spectral features to provide information on the physiological state of vegetation. The normalized band depth index (NBDI), derived from continuum-removed spectra in the region of the red-edge, showed the highest potential to differentiate between sites in this study. The study thus demonstrated the capability of hyperspectral remote sensing of vegetation canopies in identifying the site factors that affect growth of E. grandis in KwaZulu Natal, South Africa.     

Segmented canonical discriminant analysis of in situ hyperspectral data for identifying 13 urban tree species

A total of 458 in situ hyperspectral data were collected from 13 urban tree species in the City of Tampa, FL, USA using a spectrometer. The 13 species include 11 broadleaf and two conifer species. Three different techniques, segmented canonical discriminant analysis (CDA), segmented principal component analysis (PCA) and segmented stepwise discriminate analysis (SDA), were applied and compared for dimension reduction and feature extraction. With each of the three techniques, 10 features were extracted or selected from four spectral regions, visible (VIS: 1412–1797 nm), near-infrared (NIR: 707–1352 nm), mid-infrared 1 (MIR1: 1412–1797 nm) and mid-infrared 2 (MIR2: 1942–2400 nm), and used to discriminate the 13 urban tree species with a linear discriminate analysis (LDA) method. The cross-validation results, based on training samples that were used in the feature reduction step, and the results calculated from the test samples were used for evaluating the ability of the in situ hyperspectral data and performance of the segmented CDA, PCA and SDA to identify the 13 tree species. The experimental results indicate that a satisfactory discrimination of the 13 tree species was achieved using the segmented CDA technique (average accuracy (AA) = 96%, overall accuracy (OAA) = 96% and kappa = 0.958 from the cross-validation results; AA = 90%, OAA = 90% and kappa = 0.896 from the test samples) compared to the segmented PCA and SDA techniques, respectively (AA = 76% and 86%, OAA = 78% and 87%, and kappa = 0.763 and 0.857 from the cross-validation results; AA = 79% and 88%, OAA = 80% and 89%, and kappa = 0.782 and 0.879 from the test samples). In this study, the segmented CDA transformation is effective for dimension reduction and feature extraction for species discrimination with a relatively limited number of training samples. It outperformed the segmented PCA and SDA methods and produced the highest accuracies. The NIR and MIR1 regions have greater power for identifying the 13 species compared to the VIS and MIR2 spectral regions. The results indicate that CDA or segmented CDA could be applied broadly in mapping forest cover types, species identification and/or other land use/land cover classification practices with hyperspectral remote sensing data.     

Mapping wetlands using ASTER data: a comparison between classification trees and logistic regression

This study compared a non‐parametric and a parametric model for discriminating among uplands (non‐wetlands), woody wetlands, emergent wetlands and open water. Satellite images obtained on 6 March 2005 and 16 October 2005 from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) and geographic information system (GIS) data layers formed the input for analysis using classification and regression tree (CART^®) and multinomial logistic regression analysis. The overall accuracy of the CART model was 73.3%. The overall accuracy of the logit model was 76.7%. The accuracies were not statistically different from each other (McNemar χ ² = 1.65, p = 0.19). The CART producer's accuracy of the emergent wetlands was higher than the accuracy from the multinomial logit (57.1% vs. 40.7%), whereas woody wetlands identified by the multinomial logit model presented a producer's accuracy higher than that from the CART model (68.7% vs. 52.6%). A McNemar test between the two models and National Wetland Inventory (NWI) maps showed that their accuracies were not statistically different. Overall, these two models provided promising results, although they are not sufficiently accurate to replace current methods of wetland mapping based on feature extraction in high‐resolution orthoimagery.     

Predicting the big five personality traits from hand-written text features through semi-supervised learning

Handwriting analysis is a systematic study of preserved graphic structures. Which are generated in the human brain and produced on paper in cursive or printed style. The style in which a text is written reflects an array of meta-information. Personality is a combination of an individual’s behavior, emotion, motivation, and thought-pattern characteristics. It has an impact on one’s life choices, well-being, health, and numerous other preferences. This study investigates the correlation between handwriting features and personality characteristics. The prediction of personality through handwriting analysis needs to investigate the style and structure of writing. This study extracts eleven features from handwriting samples using a graph-based writing representation approach. The Big Five model of personality traits is utilized to find the personality of the writer. To improve classification accuracy utilizes a Semi-supervised Generative Adversarial Network (SGAN). This network uses a small amount of labeled data and a larger amount of unlabeled data to train the classifier. The discriminator works as a multi-class classifier and is trained on labeled, unlabeled, and generator created data. The proposed system predicts 91.3% correct personality results by utilizing the writing features of 173 participants.

     

安徽省利辛县平原区人工林树种分类研究

快速准确地绘制平原区人工林树种分布对研究人工林的生态水文和社会经济效益具有重要的意义。将资源3号（ZY-3）全色波段分别同ZY-3多光谱、哨兵2号多光谱进行融合,在图像分割基础上提取变量,采用分层优化变量组合的随机森林分类方法对安徽省利辛县人工林树种进行分类,并与分类回归树和随机森林相比较。结果表明：利用分层分类方法,平原区的人工林树种分类精度可以达到92%以上;基于哨兵2号和ZY-3融合的光谱特征变量分类精度比ZY-3数据本身的融合提高了2.49%～2.91%;而分别加入纹理变量后,分层分类方法大幅度提高了树种分类精度。因此,基于高分辨率遥感数据的光谱和纹理特征,采用分层分类方法,可以有效提高平原区人工林树种的分类精度。     

Classifying species of individual trees by intensity and structure features derived from airborne laser scanner data

The objective of this study was to identify candidate features derived from airborne laser scanner (ALS) data suitable to discriminate between coniferous and deciduous tree species. Both features related to structure and intensity were considered. The study was conducted on 197 Norway spruce and 180 birch trees (leaves on conditions) in a boreal forest reserve in Norway. The ALS sensor used was capable of recording multiple echoes. The point density was 6.6 m^− 2. Laser echoes located within the vertical projection of the tree crowns, which were assumed to be circular and defined according to field measurements, were attributed to three categories: “first echoes of many”, “single echoes”, or “last echoes of many echoes”. They were denoted FIRST, SINGLE, and LAST, respectively. In tree species classification using ALS data features should be independent of tree heights. We found that many features were dependent on tree height and that this dependency influenced selection of candidate features. When we accounted for this dependency, it was revealed that FIRST and SINGLE echoes were located higher and LAST echoes lower in the birch crowns than in spruce crowns. The intensity features of the FIRST echoes differed more between species than corresponding features of the other echo categories. For the FIRST echoes the intensity values tended to be higher for birch than spruce. When using the various features for species classification, maximum overall classification accuracies of 77% and 73% were obtained for structural and intensity features, respectively. Combining candidate features related to structure and intensity resulted in an overall classification accuracy of 88%.     

Discriminating the papyrus vegetation (Cyperus papyrus L.) and its co-existent species using random forest and hyperspectral data resampled to HYMAP

Techniques for discriminating swamp wetland species are critical for the rapid assessment and proactive management of wetlands. In this study, we tested whether the random forest (RF) algorithm could discriminate between papyrus swamp and its co-existent species (Phragmites australis, Echinochloa pyramidalis and Thelypteris interrupta) using in situ canopy reflectance spectra. Canopy spectral measurements were taken from the species using analytical spectral devices but later resampled to Hyperspectral Mapper (HYMAP) resolution. The RF algorithm and a simple forward variable selection (FVS) technique were used to identify key wavelengths for discriminating papyrus swamp and its co-existence species. The method yielded 10 wavelengths located in the visible and short-wave infrared portions of the electromagnetic spectrum with a lowest out-of-bag (OOB) estimate error rate of 9.5% and .632+?bootstrap error of 8.95%. The use of RF as a classification algorithm resulted in overall accuracy of 90.50% and a kappa value of 0.87, with individual class accuracies ranging from 93.73% to 100%. Additionally, the results from this study indicate that the RF algorithm produces better classification results than conventional classification trees (CTs) when using all HYMAP wavelengths (n?=?126) and when using wavelengths selected by the FVS technique.     

Dynamic analysis of Robinia pseudoacacia forest health levels from 1995 to 2013 in the Yellow River Delta,China using multitemporal Landsat imagery

The Robinia pseudoacacia forest in the Yellow River Delta (YRD), China, was planted in the 1970s and has continuously suffered dieback and mortality since the 1990s. Timely and accurate information on forest growth and forest condition and its dynamic change as well is essential for assessing and developing effective management strategies. In this study, multitemporal Landsat imagery was used to analyze and monitor changes of the R. pseudoacacia forest in the YRD from 1995 to 2013. To do so, Landsat image band reflectance, three fraction images calculated by using a multiple endmember spectral mixture analysis (MESMA) method, and four vegetation indices (VIs) were used to discriminate three health levels of R. pseudoacacia forest in years 1995, 2007, and 2013 with a random forest (RF) classifier. The four VIs include a difference infrared index (DII) developed in this study, normalized difference vegetation index, soil-adjusted vegetation index, and normalized difference infrared index (NDII), all of which were computed from Landsat Thematic Mapper and Operational Land Imager multispectral (MS) bands. The dynamic changes of the forest health levels during the periods of 1995–2007 and 2007–2013 were analysed. The analysis results demonstrate that three fraction images created by MESMA method and four VIs were powerful in separating the three forest health levels. In addition to the Landsat MS bands, the additional three fraction images increased the classification accuracy by 14?20%; if coupled with the four VIs, the overall accuracy was further increased by 5?6%. According to the importance values calculated by RF classifier for all input features, the DII vegetation index was the second effective feature, outperforming NDII. From 1995 to 2013, a total of 2615 ha of forest in the study area suffered from mortality or loss.     

A comparison of machine learning approaches for identifying high-poverty counties: robust features of DMSP/OLS night-time light imagery

The goal of the present study is to demonstrate that high-poverty counties and robust classification features can be identified by machine learning approaches using only DMSP/OLS night-time light imagery. To accomplish this goal, a total of 96 high-poverty and 96 non-poverty counties were classified using 15 statistical and spatial features extracted from night-time light imagery in China in 2010 formed a training set for identifying high-poverty counties. Seven machine learning approaches were adopted to classify high-poverty counties, and five feature importance measures were used to select robust features. The resulting metrics, including the user’s (>63%), producer’s (>66%) and overall (>82%) accuracies of the poor county identification (probability of poverty greater than 0.6), show that the seven machine learning approaches used in this paper exhibit good performance, although some differences exist among the approaches. The order of feature importance reveals that the relative importance of each feature differs among the models; however, the important features remain consistent. The nine most important features ranked in each approach are relatively robust for poverty identification at the county level. Both spatial feature and statistical features calculated in part from the central tendency, degree of dispersion, and the distribution of the night-time light data were identified as indispensable robust features in all the approaches, indicating that the complex social phenomenon of poverty requires analysis from different aspects. Previous studies that utilized primarily night-time light imagery applied single features related to the central tendency or the distribution features of the imagery; this study provides a new method and can act as a reference for feature selection and identification of high-poverty counties using night-time light imagery and has potential applications across several scientific domains.