Remote Sensing

Gravity data have been frequently used in researching the subsurface to map the 3D geometry of the density structure, which is considered the basis for further interpretations, such as the estimation of exploration potential in mineral exploration. The gravity inversion, practically employed to map the density structure, can be achieved by different methods. The method based on Tikhonov regularization is the most commonly used among them. Usually, the subsurface is discretized into a set of cells or voxels. To recover a stable and reliable solution, constraints are introduced into the Tikhonov regularization. One constrained inversion introduces a quadratic penalty (L₂ norm) into the inversion, which imposes smooth features on the recovered model. Another gravity inversion, known as sparse inversion, imposes compactness and sharp boundaries on the recovered density structure. Specifically, the L₁ norm and L₀ norm are favored for such a purpose. This work evaluates the merits of the gravity data inversion in cooperation with different model norms and their applicability in exploration potential estimation. Because these norms promote different features in the recovered models, the reconstructed 3D density structure reveals different geometric features of the ore deposit. We use two types of synthetic data for evaluating the performances of the inversion with different norms. Numerical results demonstrate that L₀ norm-based inversion provides high-resolution recovered models and offers reliable estimates of exploration potential with minimal deviation from theoretical mass compared to inversions equipped with the other two norms. Finally, we use the gravity data collected over the iron ore deposit at the Dida mining area in Jilin province (Northeast China) for the application. It is estimated that the exploration potential of the iron ore deposits is about 3.2 million tons. Full article

Coal is one of the fundamental fossil energy supporting the world’s economy. The synergistic development between efficient coal mining and ecological environment protection is the inevitable requirement for the preservation of global harmony. As the world’s largest coal producer, China has conducted a strategic shift from east to west in terms of the exploitation of its energy resources, posing a serious threat to the fragile ecological environment of the western region. In particular, the surface subsidence caused by coal mining is the root of the ecological deteriotation and the destruction of ground structures. However, it is difficult to reveal the law of large-scale surface subsidence in western mining areas merely by conventional measurement methods such as leveling, on account of the high intensity of coal seam mining, the weakness of the lithology of overlying rock and the large thickness of wind-blown sand strata. In view of this, small baseline subset interferometric synthetic aperture radar (SBAS-InSAR) technology was used in this study to obtain the time series of surface vertical displacement during the whole mining process of the 2401 working face in the Yingpanhao coal mine, Inner Mongolia. Based on the deformation data, the dynamic evolution characteristics of surface subsidence under high intensity mining in the western mining area were analyzed exhaustively. It was found that the surface subsidence is characterized by an extensive coverage range (48.52 km²) with minimal ground settlement (250 mm) in the study area. Meanwhile, the boundary shape of the subsidence basin followed a “circular-parallelogram-trapezoid” changeable process and the coverage area of the basin experienced three stages: a linear increasing period, a temporary stagnation period, and a re-expansion period. Furthermore, there existed an abnormal uplift phenomenon on the east side of the open-off cut in the 2401 working face. Combined with the structure of overlying strata, this paper carried out a preliminary analysis on the reasons of the abovementioned phenomenon. The research results are of vital realistic significance for ground buildings and ecological environmental protection in the aeolian sand mining area in Western China. Full article

Archaeological research is increasingly embedding individual sites in archaeological contexts and aims at reconstructing entire historical landscapes. In doing so, it benefits from technological developments in the field of archaeological prospection over the last 20 years, including LiDAR-based Digital Terrain Models, special visualizations, and automated site detection. The latter can generate comprehensive datasets with manageable effort that are useful for answering large-scale archaeological research questions. This article presents a highly automated workflow, in which a Convolutional Neural Network is used to detect burial mounds in the proximity of remotely located hollow ways. Detected mounds are then analyzed with respect to their distribution and a possible spatial relation to hollow ways. The detection works well, produces a reasonable number of results, and achieved a precision of at least 77%. The distribution of mounds shows a clear maximum in the radius of 2000–2500 m. This supports future research such as visibility or cost path analysis. Full article

The accurate and rapid estimation of the aboveground biomass (AGB) of rice is crucial to food security. Unmanned aerial vehicles (UAVs) mounted with hyperspectral sensors can obtain images of high spectral and spatial resolution in a quick and effective manner. Integrating UAV-based spatial and spectral information has substantial potential for improving crop AGB estimation. Hyperspectral remote-sensing data with more continuous reflectance information on ground objects provide more possibilities for band selection. The use of band selection for the spectral vegetation index (VI) has been discussed in many studies, but few studies have paid attention to the band selection of texture features in rice AGB estimation. In this study, UAV-based hyperspectral images of four rice varieties in five nitrogen treatments (N0, N1, N2, N3, and N4) were obtained. First, multiple spectral bands were used to identify the optimal bands of the spectral vegetation indices, as well as the texture features; next, the vegetation index model (VI model), the vegetation index combined with the corresponding-band textures model (VI+CBT model), and the vegetation index combined with the full-band textures model (VI+FBT model) were established to compare their respective rice AGB estimation abilities. The results showed that the optimal bands of the spectral and textural information for AGB monitoring were inconsistent. The red-edge and near-infrared bands demonstrated a strong correlation with the rice AGB in the spectral dimension, while the green and red bands exhibited a high correlation with the rice AGB in the spatial dimension. The ranking of the monitoring accuracies of the three models, from highest to lowest, was: the VI+FBT model, then the VI+CBT model, and then the VI model. Compared with the VI model, the R² of the VI+FBT model and the VI+CBT model increased by 1.319% and 9.763%, respectively. The RMSE decreased by 2.070% and 16.718%, respectively, while the rRMSE decreased by 2.166% and 16.606%, respectively. The results indicated that the integration of vegetation indices and textures can significantly improve the accuracy of rice AGB estimation. The full-band textures contained richer information that was highly related to rice AGB. The VI model at the tillering stage presented the greatest sensitivity to the integration of textures, and the models in the N3 treatment (1.5 times the normal nitrogen level) gave the best AGB estimation compared with the other nitrogen treatments. This research proposes a reliable modeling framework for monitoring rice AGB and provides scientific support for rice-field management. Full article

Near-surface ozone is one of the significant issues in the troposphere. Recently, ozone pollution in Lhasa at an altitude of 3600 m has caused attention. The current knowledge of ozone formation in Lhasa city is still minimal. In this work, the profile of VOCs during early summer was investigated, and alkanes were the most critical group of VOCs. The urban areas of Lhasa are under transition conditions and controlled by both VOCs and NOx. Moreover, the most effective way to decrease ozone formation is to reduce the emissions of anthropogenic VOCs and NOx. The vertical distribution of tropospheric ozone was investigated using differential absorption lidar (DIAL). The results show that ozone concentrations decreased with the elevation of altitudes over Lhasa. The vertical distribution showed clear diurnal trends and that a high ozone concentration appeared at night because of the afternoon’s accumulated O₃ generated by photochemical reactions. Ozone in Lhasa is mainly distributed between 0.4 km and 0.6 km. Local generation, overnight accumulation, and NOx titration were identified as three ozone distribution modes. This work helps to understand ozone formation and distribution in the Tibetan Plateau. Full article

When a natural disaster occurs, humanitarian organizations need to be prompt, effective, and efficient to support people whose security is threatened. Satellite imagery offers rich and reliable information to support expert decision-making, yet its annotation remains labour-intensive and tedious. In this work, we evaluate the applicability of convolutional neural networks (CNN) in supporting building damage assessment in an emergency context. Despite data scarcity, we develop a deep learning workflow to support humanitarians in time-constrained emergency situations. To expedite decision-making and take advantage of the inevitable delay to receive post-disaster satellite images, we decouple building localization and damage classification tasks into two isolated models. Our contribution is to show the complexity of the damage classification task and use established transfer learning techniques to fine-tune the model learning and estimate the minimal number of annotated samples required for the model to be functional in operational situations. Full article

In this article, we propose a method for automatic calibration of a LiDAR–camera system, which can be used in autonomous cars. This approach does not require any calibration pattern, as calibration is only based on real traffic scenes observed by sensors; the results of camera image segmentation are compared with scanning LiDAR depth data. The proposed algorithm superimposes the edges of objects segmented by the Mask-RCNN network with depth discontinuities. The method can run in the background during driving, and it can automatically detect decalibration and correct corresponding rotation matrices in an online and near real-time mode. Experiments on the KITTI dataset demonstrated that, for input data of moderate quality, the algorithm could calculate and correct rotation matrices with an average accuracy of 0.23°. Full article

The Northeast China cold vortex (NCCV) often occurs in spring and summer, causing extreme weather such as rainstorm and hail in Northeast China. The brightness temperature (TB) observations of Microwave Temperature Sounder-2 (MWTS-2) on board Fengyun-3D (FY-3D), which can provide atmospheric temperature in various vertical layers, are firstly limb-corrected and then applied to track the origin and movement of four NCCV cases in June and July 2019. Results show that a cold core is observed at the location of NCCVs in TB observations of channels 4 and 5, whose peak weighting function (WF) altitudes are 700 and 400 hPa, respectively, indicating the cold structure of NCCVs in the middle and lower troposphere. The TB observations of channels 6 and 7, which measure the atmospheric temperature around 250 and 200 hPa, respectively, capture a warm core structure of NCCVs in the upper troposphere and lower stratosphere. Being less affected by the low-level cloud and rain, TB observations of channels 6 and 7 are applied to identify and track the warm cores of NCCVs. The NCCV tracks of movement identified by MWTS-2 observations compare well with those determined by the 500 hPa geopotential height and the 300 hPa potential vorticity (PV) anomaly from the ERA5 reanalysis. Both clearly show that the NCCVs were originated from high latitudes, then moved southeastward, and finally entered Northeast China. The entire process took several days. Therefore, TB observations of MWTS-2 can be used to identify the precursors of NCCVs and monitor their appearances, developments, and movements in time. With the flourishing development of Fengyun satellite series in China as well as the already existing 40 years of microwave sounder observations worldwide, this research provides a new way to investigate the synoptic and climatological features of NCCVs. Full article

Beijing is a major city suffering from land subsidence due to long-term over-exploitation of groundwater. The South-to-North Water Diversion Project (SNWDP), however, has had a significant impact on the structure of water consumption since the end of 2014, and it is changing the status of land subsidence in Beijing. In this study, we employed Persistent Scatterers Synthetic Aperture Radar Interferometric (PS-InSAR) to investigate the decadal evolution of land subsidence in Beijing with 100 TerraSAR-X stripmap images collected from April 2010 to December 2019. The water resources, historic climate and urban construction data were compiled for the years of 2010 to 2019 to analyze changes in groundwater level, human activity, surface geology, active faults and land subsidence patterns. The results show that the changes in the water supply structure are correlated to a rise in groundwater level after 2015. These changes include an increase in the water supply from the SNWDP, a reduction in groundwater exploitation, the optimization of water consumption, replacing recycled water for environmental water and a reduction in the use of water for agriculture. Land subsidence in the study area was concentrated in the eastern regions, trending towards a decreasing velocity starting about two years after the commencement of SNWDP in 2015. Uneven subsidence in the land subsidence area was related to excavations of underground soil, and the construction of Line 6 and Line 7 led to rapid nonlinear subsidence. Our results have scientific significance for reducing subsidence hazards in the context of SNWDP and urban expansion. Full article

Large-scale open burning of straw residues causes seasonal and severe atmospheric pollution in Northeast China. Previous studies focused on the causes or assessment of atmospheric pollution in a single city. However, studies conducted on the interaction range, degree and policy control of pollutant transport on a large scale are still to be performed. In this study, we propose combined control of straw burning by dividing region the straw burning in Northeast China in recent 20 years, determining the transport routes between main cities, and analyzing the interaction characteristics of straw burning under different scenarios. The fire point data suggest that the most intense straw burning years in Northeast China in the past 20 years occurred in the range from 2014 to 2017, mainly after the autumn harvest (October–November) and before spring cultivation (March–April). The burning areas were concentrated in the belt of Shenyang-Changchun-Harbin, the border of the three provinces and Eastern-Inner Mongolia, and the surrounding area of Hegang and Jiamusi City. The lower number of fire points before 2013 indicates that high-intensity burning has not always been the case, while the sharp decline after 2018 is mainly due to scientific control of straw burning and increased comprehensive utilization of straw. Compared with S2, the PM_2.5 concentrations increased by 6.2% in S3 and 18.7% in S4, indicating that burning in three or four provinces at the same time will significantly increase air pollution and exert a regional transmission effect. Straw burning in Northeast China is divided into six main regions based on correlation analysis and satellite fire monitoring. Under typical S3, the case analysis results indicate that there is regional transmission interaction between different cities and provinces, focusing on multi-province border cities, and it is affected by Northwest long airflow, and Southeast and Northeast short airflow. These results provide scientific and technological support for implementing the joint prevention and control plan for straw incineration in Northeast China. Full article

Heterogeneous synthetic aperture radar (SAR) images contain more complementary information compared with homologous SAR images; thus, the comprehensive utilization of heterogeneous SAR images could potentially improve performance for the monitoring of sea surface objects, such as sea ice and enteromorpha. Image registration is key to the application of monitoring sea surface objects. Heterogeneous SAR images have intensity differences and resolution differences, and after the uniform resolution, intensity differences are one of the most important factors affecting the image registration accuracy. In addition, sea surface objects have numerous repetitive and confusing features for feature extraction, which also limits the image registration accuracy. In this paper, we propose an improved L2Net network for image registration with intensity differences and repetitive texture features, using sea ice as the research object. The deep learning network can capture feature correlations between image patch pairs, and can obtain the correct matching from a large number of features with repetitive texture. In the SAR image pair, four patches of different sizes centered on the corner points are proposed as inputs. Thus, local features and more global features are fused to obtain excellent structural features, to distinguish between different repetitive textural features, add contextual information, further improve the feature correlation, and improve the accuracy of image registration. An outlier removal strategy is proposed to remove false matches due to repetitive textures. Finally, the effectiveness of our method was verified by comparative experiments. Full article

Evapotranspiration ($ET$) connects the land to the atmosphere, linking water, energy, and carbon cycles. $ET$ is an essential climate variable with a fundamental importance, and accurate assessments of the spatiotemporal trends and variability in $ET$ are needed from regional to continental scales. This study compared eight global actual $ET$ datasets ($E{T}_{gl}$) and the average actual $ET$ ensemble ($E{T}_{ens}$) based on remote sensing, climate reanalysis, land-surface, and biophysical models to $ET$ computed from basin-scale water balance ($E{T}_{wb}$) in South America on monthly time scale. The 50 small-to-large basins covered major rivers and different biomes and climate types. We also examined the magnitude, seasonality, and interannual variability of $ET$, comparing $E{T}_{gl}$ and $E{T}_{ens}$ with $E{T}_{wb}$. Global $ET$ datasets were evaluated between 2003 and 2014 from the following datasets: Breathing Earth System Simulator (BESS), ECMWF Reanalysis 5 (ERA5), Global Land Data Assimilation System (GLDAS), Global Land Evaporation Amsterdam Model (GLEAM), MOD16, Penman–Monteith–Leuning (PML), Operational Simplified Surface Energy Balance (SSEBop) and Terra Climate. By using $E{T}_{wb}$ as a basis for comparison, correlation coefficients ranged from 0.45 (SSEBop) to 0.60 ($E{T}_{ens}$), and RMSE ranged from 35.6 ($E{T}_{ens}$) to 40.5 mm·month⁻¹ (MOD16). Overall, $E{T}_{gl}$ estimates ranged from 0 to 150 mm·month⁻¹ in most basins in South America, while $E{T}_{wb}$ estimates showed maximum rates up to 250 mm·month⁻¹. $E{T}_{gl}$ varied by hydroclimatic regions: (i) basins located in humid climates with low seasonality in precipitation, including the Amazon, Uruguay, and South Atlantic basins, yielded weak correlation coefficients between monthly $E{T}_{gl}$ and $E{T}_{wb}$, and (ii) tropical and semiarid basins (areas where precipitation demonstrates a strong seasonality, as in the São Francisco, Northeast Atlantic, Paraná/Paraguay, and Tocantins basins) yielded moderate-to-strong correlation coefficients. An assessment of the interannual variability demonstrated a disagreement between $E{T}_{gl}$ and $E{T}_{wb}$ in the humid tropics (in the Amazon), with $E{T}_{gl}$ showing a wide range of interannual variability. However, in tropical, subtropical, and semiarid climates, including the Tocantins, São Francisco, Paraná, Paraguay, Uruguay, and Atlantic basins (Northeast, East, and South), we found a stronger agreement between $E{T}_{gl}$ and $E{T}_{wb}$ for interannual variability. Assessing $ET$ datasets enables the understanding of land–atmosphere exchanges in South America, to improvement of $ET$ estimation and monitoring for water management. Full article

Extrinsic calibration on a LiDAR-camera system is an essential task for the advanced perception application for the intelligent vehicle. In the offline situation, a calibration object based method can estimate the extrinsic parameters in high precision. However, during the long time application of LiDAR-camera system in the actual scenario, the relative pose of LiDAR and camera has small and accumulated drift, so that the offline calibration result is not accurate. To correct the extrinsic parameter conveniently, we present a deep learning based online extrinsic calibration method in this paper. From Lambertian reflection model, it is found that an object with higher LiDAR intensity has the higher possibility to have salient RGB features. Based on this fact, we present a LiDAR intensity attention based backbone network (LIA-Net) to extract the significant co-observed calibration features from LiDAR data and RGB image. In the later stage of training, the loss of extrinsic parameters changes slowly, causing the risk of vanishing gradient and limiting the training efficiency. To deal with this issue, we present the structural consistency (SC) loss to minimize the difference between projected LiDAR image (i.e., LiDAR depth image, LiDAR intensity image) and its ground truth (GT) LiDAR image. It aims to accurately align the LiDAR point and RGB pixel. With LIA-Net and SC loss, we present the convolution neural network (CNN) based calibration network LIA-SC-Net. Comparison experiments on a KITTI dataset demonstrate that LIA-SC-Net has achieved more accurate calibration results than state-of-the-art learning based methods. The proposed method has both accurate and real-time performance. Ablation studies also show the effectiveness of proposed modules. Full article

Four-band color imaging of the Martian surface using the Color and Stereo Surface Imaging System (CaSSIS) onboard the European Space Agency’s ExoMars Trace Gas Orbiter exhibits a high color diversity in specific regions. Not only is the correlation of color diversity maps with local morphological properties desirable, but mineralogical interpretation of the observations is also of great interest. The relatively high spatial resolution of CaSSIS data mitigates its low spectral resolution. In this paper, we combine the broad-band imaging of the surface of Mars, acquired by CaSSIS with hyperspectral data from the Compact Reconnaissance Imaging Spectrometer (CRISM) onboard NASA’s Mars Reconnaissance Orbiter to achieve a fusion of both datasets. We achieve this using dimensionality reduction and data clustering of the high dimensional datasets from CRISM. In the presented research, CRISM data from the Coprates Chasma region of Mars are tested with different machine learning methods and compared for robustness. With the help of a suitable metric, the best method is selected and, in a further step, an optimal cluster number is determined. To validate the methods, the so-called “summary products” derived from the hyperspectral data are used to correlate each cluster with its mineralogical properties. We restrict the analysis to the visible range in order to match the generated clusters to the CaSSIS band information in the range of 436–1100 nm. In the machine learning community, the so-called UMAP method for dimensionality reduction has recently gained attention because of its speed compared to the already established t-SNE. The results of this analysis also show that this method in combination with the simple K-Means outperforms comparable methods in its efficiency and speed. The cluster size obtained is between three and six clusters. Correlating the spectral cluster maps with the given summary products from CRISM shows that four bands, and especially the NIR bands and VIS albedo, are sufficient to discriminate most of these clusters. This demonstrates that features in the four-band CaSSIS images can provide robust mineralogical information, despite the limited spectral information using semi-automatic processing. Full article

Hyperspectral remote sensing image (HSI) include rich spectral information that can be very beneficial for change detection (CD) technology. Due to the existence of many mixed pixels, pixel-wise approaches can lead to considerable errors in the resulting CD map. The spectral unmixing (SU) method is a potential solution to this problem, as it decomposes mixed pixels into a set of fractions of land cover. Subsequently, the CD map is created by comparing the abundance images. However, based only on the abundance images created through the SU method, they are unable to effectively provide detailed change information. Meanwhile, the features of change information cannot be sufficiently extracted by the traditional sub-pixel CD framework, which leads to a poor CD result. To address these problems, this paper presents an integrated CD method based on multi-endmember spectral unmixing, joint matrix and CNN (MSUJMC) for HSI. Three main steps are considered to accomplish this task. First, considering the endmember spectral variability, more reliable endmember abundance information is obtained by multi-endmember spectral unmixing (MSU). Second, the original image features are incorporated with the abundance images using a joint matrix (JM) algorithm to provide more temporal and spatial land cover change information characteristics. Third, to efficiently extract the change features and to better handle the fused multi-source information, the convolutional neural network (CNN) is introduced to realize a high-accuracy CD result. The proposed method has been verified on simulated and real multitemporal HSI datasets, which provide multiple changes. Experimental results verify the effectiveness of the proposed approach. Full article

Roadside LiDAR has become an important sensor for the detection of objects in cities, such as vehicles and pedestrians, which is due to its advantages of all-weather operation and high-ranging accuracy. In order to serve an intelligent transportation system, the efficient and accurate segmentation of vehicles and pedestrians is needed in the coverage area of the LiDAR. In this study, a roadside LiDAR was fixed on brackets on both sides of the road to obtain the point-cloud information on the urban road and the surrounding environment. A segmentation method that is based on a scanning LiDAR sensor is proposed. First, a polar grid that is based on polar coordinates is constructed to count the LiDAR rotations to obtain the original information of the angle and the distance of the point cloud, and the background point-cloud image is dynamically updated over time. By aiming at the complex urban road environment and the interference of trees and light poles in the background, an adaptive polar-grid Gaussian-mixture model (APG-GMM) that uses a point-cloud method is proposed to improve the accuracy of the foreground and background segmentation. A density-adaptive DBSCAN target-clustering algorithm is proposed, as well as a dynamic adaptive neighborhood radius, to solve the problem of the low clustering accuracy that is caused by the uneven density of point clouds that are collected by LiDAR, and to divide the point clouds in the foreground into vehicles and pedestrians. Finally, the method was tested at intersections and urban roads with dense traffic flows. The experimental results show that the proposed algorithm can segment the foreground and background well and can cluster vehicles and pedestrians while reducing the number of calculations and the time complexity. Full article

We evaluated the coupled dynamics of vegetation dynamics (NDVI) and soil moisture (SMOS) at monthly resolution over different regions of tropical South America and the effects of the Eastern Pacific (EP) and the Central Pacific (CP) El Niño–Southern Oscillation (ENSO) events. We used linear Pearson cross-correlation, wavelet and cross wavelet analysis (CWA) and three nonlinear causality methods: ParrCorr, GPDC and PCMCIplus. Results showed that NDVI peaks when SMOS is transitioning from maximum to minimum monthly values, which confirms the role of SMOS in the hydrological dynamics of the Amazonian greening up during the dry season. Linear correlations showed significant positive values when SMOS leads NDVI by 1–3 months. Wavelet analysis evidenced strong 12- and 64-month frequency bands throughout the entire record length, in particular for SMOS, whereas the CWA analyses indicated that both variables exhibit a strong coherency at a wide range of frequency bands from 2 to 32 months. Linear and nonlinear causality measures also showed that ENSO effects are greater on SMOS. Lagged cross-correlations displayed that western (eastern) regions are more associated with the CP (EP), and that the effects of ENSO manifest as a travelling wave over time, from northwest (earlier) to southeast (later) over tropical South America and the Amazon River basin. The ParrCorr and PCMCIplus methods produced the most coherent results, and allowed us to conclude that: (1) the nonlinear temporal persistence (memory) of soil moisture is stronger than that of NDVI; (2) the existence of two-way nonlinear causalities between NDVI and SMOS; (3) diverse causal links between both variables and the ENSO indices: CP (7/12 with ParrCorr; 6/12 with PCMCIplus), and less with EP (5/12 with ParrCorr; 3/12 with PCMCIplus). Full article

Blind image deblurring is a long-standing challenge in remote sensing image restoration tasks. It aims to recover a latent sharp image from a blurry image while the blur kernel is unknown. To solve this problem, many image priors-based algorithms and learning-based algorithms have been proposed. However, most of these methods are based on a single blurry image. Due to the lack of high frequency information, the images restored by these algorithms still have some deficiencies in edge and texture details. In this work, we propose a novel deep learning model named Reference-Based Multi-Level Features Fusion Deblurring Network (Ref-MFFDN), which registers the reference image and the blurry image in the multi-level feature space and transfers the high-quality textures from registered reference features to assist image deblurring. Comparative experiments on the testing set prove that our Ref-MFFDN outperforms many state-of-the-art single image deblurring approaches in both quantitative evaluation and visual results, which indicates the effectiveness of using reference images in remote sensing image deblurring tasks. More ablation experiments demonstrates the robustness of Ref-MFFDN to the input image size, the effectiveness of multi-level features fusion network (MFFN) and the effect of different feature levels in multi-feature extractor (MFE) on algorithm performance. Full article

In many parts of the world, apple trees suffer from severe foliar damage each year due to infection of Alternaria blotch (Alternaria alternata f. sp. Mali), resulting in serious economic losses to growers. Traditional methods for disease detection and severity classification mostly rely on manual labor, which is slow, labor-intensive and highly subjective. There is an urgent need to develop an effective protocol to rapidly and accurately evaluate disease severity. In this study, DeeplabV3+, PSPNet and UNet were used to assess the severity of apple Alternaria leaf blotch. For identifications of leaves and disease areas, the dataset with a total of 5382 samples was randomly split into 74% (4004 samples) for model training, 9% (494 samples) for validation, 8% (444 samples) for testing and 8% (440 samples) for overall testing. Apple leaves were first segmented from complex backgrounds using the deep-learning algorithms with different backbones. Then, the recognition of disease areas was performed on the segmented leaves. The results showed that the PSPNet model with MobileNetV2 backbone exhibited the highest performance in leaf segmentation, with precision, recall and MIoU values of 99.15%, 99.26% and 98.42%, respectively. The UNet model with VGG backbone performed the best in disease-area prediction, with a precision of 95.84%, a recall of 95.54% and a MIoU value of 92.05%. The ratio of disease area to leaf area was calculated to assess the disease severity. The results showed that the average accuracy for severity classification was 96.41%. Moreover, both the correlation coefficient and the consistency correlation coefficient were 0.992, indicating a high agreement between the reference values and the value that the research predicted. This study proves the feasibility of rapid estimation of the severity of apple Alternaria leaf blotch, which will provide technical support for precise application of pesticides. Full article