Search

High-precision landcover classification is a fundamental prerequisite for resource and environmental monitoring and land-use status surveys. Imbued with intricate spatial information and texture features, very high spatial resolution remote sensing images accentuate the divergence between features within the same category, thereby amplifying the complexity of landcover classification. Consequently, semantic segmentation models leveraging deep backbone networks have emerged as stalwarts in landcover classification tasks owing to their adeptness in feature representation. However, the classification efficacy of a solitary backbone network model fluctuates across diverse scenarios and datasets, posing a persistent challenge in the construction or selection of an appropriate backbone network for distinct classification tasks. To elevate the classification performance and bolster the generalization of semantic segmentation models, we propose a novel semantic segmentation network architecture, named the aggregated backbone network (ABNet), for the meticulous landcover classification. ABNet aggregates three prevailing backbone networks (ResNet, HRNet, and VoVNet), distinguished by significant structural disparities, using a same-stage fusion approach. Subsequently, it amalgamates these networks with the Deeplabv3+ head after integrating the convolutional block attention mechanism (CBAM). Notably, this amalgamation harmonizes distinct scale features extracted by the three backbone networks, thus enriching the model’s spatial contextual comprehension and expanding its receptive field, thereby facilitating more effective semantic feature extraction across different stages. The convolutional block attention mechanism primarily orchestrates channel adjustments and curtails redundant information within the aggregated feature layers. Ablation experiments demonstrate an enhancement of no less than 3% in the mean intersection over union (mIoU) of ABNet on both the LoveDA and GID15 datasets when compared with a single backbone network model. Furthermore, in contrast to seven classical or state-of-the-art models (UNet, FPN, PSPNet, DANet, CBNet, CCNet, and UPerNet), ABNet evinces excellent segmentation performance across the aforementioned datasets, underscoring the efficiency and robust generalization capabilities of the proposed approach.

This study assesses the applicability of different-resolution multispectral remote sensing images for mapping and estimating the aboveground biomass (AGB) of Carpobrotus edulis, a prominent invasive species in European coastal areas. This study was carried out on the Cávado estuary sand spit (Portugal). The performance of three sets of multispectral images with different Ground Sample Distances (GSDs) were compared: 2.5 cm, 5 cm, and 10 cm. The images were classified using the supervised classification algorithm random forest and later improved by applying a sieve filter. Samples of C. edulis were also collected, dried, and weighed to estimate the AGB using the relationship between the dry weight (DW) and vegetation indices (VIs). The resulting regression models were evaluated based on their coefficient of determination (R2), Normalised Root Mean Square Error (NRMSE), p-value, Akaike information criterion (AIC), and the Bayesian information criterion (BIC). The results show that the three tested image resolutions allow for constructing reliable coverage maps of C. edulis, with overall accuracy values of 89%, 85%, and 88% for the classification of the 2.5 cm, 5 cm, and 10 cm GSD images, respectively. The best-performing VI-DW regression models achieved R2 = 0.87 and NRMSE = 0.09 for the 2.5 cm resolution; R2 = 0.77 and NRMSE = 0.12 for the 5 cm resolution; and R2 = 0.64 and NRMSE = 0.15 for the 10 cm resolution. The C. edulis area and total AGB were 3441.10 m2 and 28,327.1 kg (with an AGB relative error (RE) = 0.08) for the 2.5 cm resolution; 3070.04 m2 and 29,170.8 kg (AGB RE = 0.08) for the 5 cm resolution; and 2305.06 m2 and 22,135.7 kg (AGB RE = 0.11) for the 10 cm resolution. Spatial and model differences were analysed in detail to determine their causes. Final analyses suggest that multispectral imagery of up to 5 cm GSD is adequate for estimating C. edulis distribution and biomass.

Hyperspectral images (HSIs) provide rich spectral information, facilitating many applications, including landcover classification. However, due to the high dimensionality of HSIs, landcover mapping applications usually suffer from the curse of dimensionality, which degrades the efficiency of supervised classifiers due to insufficient training samples. Feature extraction (FE) is a popular dimension reduction strategy for this issue. This paper proposes an unsupervised FE algorithm that involves extracting endmembers and clustering spectral bands. The proposed method first extracts existing endmembers from the HSI data via a vertex component analysis method. Using these endmembers, it subsequently constructs a prototype space (PS) in which each spectral band is represented by a point. Similar/correlated bands in the PS remain near one another, forming several clusters. Therefore, our method, in the next step, clusters spectral bands into multiple clusters via K-means and fuzzy C-means algorithms. Finally, it combines all the spectral bands in the same cluster using a weighted average operator to decrease the high dimensionality. The extracted features were evaluated by applying an SVM classifier. The experimental results confirmed the superior performance of the proposed method compared with five state-of-the-art dimension reduction algorithms. It outperformed these algorithms in terms of classification accuracy on three widely used hyperspectral images (Indian Pines, KSC, and Pavia Centre). The suggested technique also showed comparable or even stronger performance (up to 9% improvement) compared with its supervised competitor. Notably, the proposed method exhibited higher accuracy even when only a limited number of training samples were available for supervised classification. Using only five training samples per class for the KSC and Pavia Centre datasets, our method’s classification accuracy was higher than that of its best-performing unsupervised competitors by about 7% and 1%, respectively, in our experiments.

Current Earth observation systems generate massive amounts of satellite image time series to keep track of geographical areas over time to monitor and identify environmental and climate change. Efficiently analyzing such data remains an unresolved issue in remote sensing. In classifying land cover, utilizing SITS rather than one image might benefit differentiating across classes because of their varied temporal patterns. The aim was to forecast the land cover class of a group of pixels as a multi-class single-label classification problem given their time series gathered using satellite images. In this article, we exploit SITS to assess the capability of several spatial and temporal deep learning models with the proposed architecture. The models implemented are the bidirectional gated recurrent unit (GRU), temporal convolutional neural networks (TCNN), GRU + TCNN, attention on TCNN, and attention of GRU + TCNN. The proposed architecture integrates univariate, multivariate, and pixel coordinates for the Reunion Island’s landcover classification (LCC). the evaluation of the proposed architecture with deep neural networks on the test dataset determined that blending univariate and multivariate with a recurrent neural network and pixel coordinates achieved increased accuracy with higher F1 scores for each class label. The results suggest that the models also performed exceptionally well when executed in a partitioned manner for the LCC task compared to the temporal models. This study demonstrates that using deep learning approaches paired with spatiotemporal SITS data addresses the difficult task of cost-effectively classifying land cover, contributing to a sustainable environment.

Landcover classification is an important application in remote sensing, but it is always a challenge to distinguish different features with similar characteristics or large-scale differences. Some deep learning networks, such as UperNet, PSPNet, and DANet, use pyramid pooling and attention mechanisms to improve their abilities in multi-scale features extraction. However, due to the neglect of low-level features contained in the underlying network and the information differences between feature maps, it is difficult to identify small-scale objects. Thus, we propose a novel image segmentation network, named HFENet, for mining multi-level semantic information. Like the UperNet, HFENet adopts a top-down horizontal connection architecture while includes two improved modules, the HFE and the MFF. According to the characteristics of different levels of semantic information, HFE module reconstructs the feature extraction part by introducing an attention mechanism and pyramid pooling module to fully mine semantic information. With the help of a channel attention mechanism, MFF module up-samples and re-weights the feature maps to fuse them and enhance the expression ability of multi-scale features. Ablation studies and comparative experiments between HFENet and seven state-of-the-art models (U-Net, DeepLabv3+, PSPNet, FCN, UperNet, DANet and SegNet) are conducted with a self-labeled GF-2 remote sensing image dataset (MZData) and two open datasets landcover.ai and WHU building dataset. The results show that HFENet on three datasets with six evaluation metrics (mIoU, FWIoU, PA, mP, mRecall and mF1) are better than the other models and the mIoU is improved 7.41–10.60% on MZData, 1.17–11.57% on WHU building dataset and 0.93–4.31% on landcover.ai. HFENet can perform better in the task of refining the semantic segmentation of remote sensing images.

As an important application in remote sensing, landcover classification remains one of the most challenging tasks in very-high-resolution (VHR) image analysis. As the rapidly increasing number of Deep Learning (DL) based landcover methods and training strategies are claimed to be the state-of-the-art, the already fragmented technical landscape of landcover mapping methods has been further complicated. Although there exists a plethora of literature review work attempting to guide researchers in making an informed choice of landcover mapping methods, the articles either focus on the review of applications in a specific area or revolve around general deep learning models, which lack a systematic view of the ever advancing landcover mapping methods. In addition, issues related to training samples and model transferability have become more critical than ever in an era dominated by data-driven approaches, but these issues were addressed to a lesser extent in previous review articles regarding remote sensing classification. Therefore, in this paper, we present a systematic overview of existing methods by starting from learning methods and varying basic analysis units for landcover mapping tasks, to challenges and solutions on three aspects of scalability and transferability with a remote sensing classification focus including (1) sparsity and imbalance of data; (2) domain gaps across different geographical regions; and (3) multi-source and multi-view fusion. We discuss in detail each of these categorical methods and draw concluding remarks in these developments and recommend potential directions for the continued endeavor.

The role of forests to sequester carbon is considered an important strategy for mitigating climate change and achieving net zero emissions. However, forests in North Korea have continued to be cleared since the 1990s due to the lack of food and energy resources. Deforestation in this country has not been accurately classified nor consistently reported because of the characteristics of small patches. This study precisely determined the area of deforested land in North Korea through the vegetation phenological classification using high-resolution satellite imagery and deep learning algorithms. Effective afforestation target sites in North Korea were identified with priority grade. The U-Net deep learning algorithm and time-series Sentinel-2 satellite images were applied to phenological classification; the results reflected the small patch-like characteristics of deforestation in North Korea. Based on the phenological classification, the land cover of the country was classified with an accuracy of 84.6%; this included 2.6 million ha of unstocked forest and reclaimed forest. Sites for afforestation were prioritized into five grades based on deforested characteristics, altitude and slope. Forest area is expanded and the forest ecosystem is restored through successful afforestation, this may improve the overall ecosystem services in North Korea. In the long term, it will be possible to contribute to carbon neutrality and greenhouse gas reduction on the Korean Peninsula level through optimal afforestation by using these outcomes.

Images provided by the ESA Sentinel-2 mission are rapidly becoming the main source of information for the entire remote sensing community, thanks to their unprecedented combination of spatial, spectral and temporal resolution, as well as their associated open access policy. Due to a sensor design trade-off, images are acquired (and delivered) at different spatial resolutions (10, 20 and 60 m) according to specific sets of wavelengths, with only the four visible and near infrared bands provided at the highest resolution (10 m). Although this is not a limiting factor in general, many applications seem to emerge in which the resolution enhancement of 20 m bands may be beneficial, motivating the development of specific super-resolution methods. In this work, we propose to leverage Convolutional Neural Networks (CNNs) to provide a fast, upscalable method for the single-sensor fusion of Sentinel-2 (S2) data, whose aim is to provide a 10 m super-resolution of the original 20 m bands. Experimental results demonstrate that the proposed solution can achieve better performance with respect to most of the state-of-the-art methods, including other deep learning based ones with a considerable saving of computational burden.

Land cover Classified maps obtained from deep learning methods such as Convolutional neural networks (CNNs) and fully convolutional networks (FCNs) usually have high classification accuracy but with the detailed structures of objects lost or smoothed. In this work, we develop a methodology based on fully convolutional networks (FCN) that is trained in an end-to-end fashion using aerial RGB images only as input. Skip connections are introduced into the FCN architecture to recover high spatial details from the lower convolutional layers. The experiments are conducted on the city of Goma in the Democratic Republic of Congo. We compare the results to a state-of-the art approach based on a semi-automatic Geographic object image-based analysis (GEOBIA) processing chain. State-of-the art classification accuracies are obtained by both methods whereby FCN and the best baseline method have an overall accuracy of 91.3% and 89.5% respectively. The maps have good visual quality and the use of an FCN skip architecture minimizes the rounded edges that is characteristic of FCN maps. Additional experiments are done to refine FCN classified maps using segments obtained from GEOBIA generated at different scale and minimum segment size. High OA of up to 91.5% is achieved accompanied with an improved edge delineation in the FCN maps, and future work will involve explicitly incorporating boundary information from the GEOBIA segmentation into the FCN pipeline in an end-to-end fashion. Finally, we observe that FCN has a lower computational cost than the standard patch-based CNN approach especially at inference.