Networks For Land Cover Classification Research Articles

DGFNet: Dual Gate Fusion Network for Land Cover Classification in Very High-Resolution Images

Deep convolutional neural networks (DCNNs) have been used to achieve state-of-the-art performance on land cover classification thanks to their outstanding nonlinear feature extraction ability. DCNNs are usually designed as an encoder–decoder architecture for the land cover classification in very high-resolution (VHR) remote sensing images. The encoder captures semantic representation by stacking convolution layers and shrinking image spatial resolution, while the decoder restores the spatial information by an upsampling operation and combines it with different level features through a summation or skip connection. However, there is still a semantic gap between different-level features; a simple summation or skip connection will reduce the performance of land-cover classification. To overcome this problem, we propose a novel end-to-end network named Dual Gate Fusion Network (DGFNet) to restrain the impact of the semantic gap. In detail, the key of DGFNet consists of two main components: Feature Enhancement Module (FEM) and Dual Gate Fusion Module (DGFM). Firstly, the FEM combines local information with global contents and strengthens the feature representation in the encoder. Secondly, the DGFM is proposed to reduce the semantic gap between different level features, effectively fusing low-level spatial information and high-level semantic information in the decoder. Extensive experiments conducted on the LandCover dataset and the ISPRS Potsdam dataset proved the effectiveness of the proposed network. The DGFNet achieves state-of-art performance 88.87% MIoU on the LandCover dataset and 72.25% MIoU on the ISPRS Potsdam dataset.

Domain-Adversarial Training of Self-Attention-Based Networks for Land Cover Classification Using Multi-Temporal Sentinel-2 Satellite Imagery

The increasing availability of large-scale remote sensing labeled data has prompted researchers to develop increasingly precise and accurate data-driven models for land cover and crop classification (LC&CC). Moreover, with the introduction of self-attention and introspection mechanisms, deep learning approaches have shown promising results in processing long temporal sequences in the multi-spectral domain with a contained computational request. Nevertheless, most practical applications cannot rely on labeled data, and in the field, surveys are a time-consuming solution that pose strict limitations to the number of collected samples. Moreover, atmospheric conditions and specific geographical region characteristics constitute a relevant domain gap that does not allow direct applicability of a trained model on the available dataset to the area of interest. In this paper, we investigate adversarial training of deep neural networks to bridge the domain discrepancy between distinct geographical zones. In particular, we perform a thorough analysis of domain adaptation applied to challenging multi-spectral, multi-temporal data, accurately highlighting the advantages of adapting state-of-the-art self-attention-based models for LC&CC to different target zones where labeled data are not available. Extensive experimentation demonstrated significant performance and generalization gain in applying domain-adversarial training to source and target regions with marked dissimilarities between the distribution of extracted features.

A random patches based edge preserving network for land cover classification using Polarimetric Synthetic Aperture Radar images

ABSTRACT A random patches-based edge-preserving network (RPEP) is proposed for polarimetric synthetic aperture radar (PolSAR) image classification in this paper. An initial spatial feature extraction is firstly done using the transform domain recursive filtering. From the filtered images, several random patches are chosen and used as convolutional kernels. The designed random patches-based network uses these fix kernels without doing any training process. The multi-scale features extracted by both shallow and deep layers are given to a support vector machine to get an initial classification map. The binary probability maps obtained from the initial classification map are then smoothed using the guided filter as an edge preserving filter. The final classification map is achieved by applying the maximum decision rule. The proposed RPEP method with extraction of robust, consistent, invariant and multi-scale polarimetric-spatial features and also by doing noise reduction with edge preserving filters provides superior classification results compared to several state-of-the-art methods especially in small sample size situations. In addition, RPEP has a simple and fast implementation that makes it a powerful classifier for real applications.

Sparse Pixel Training of Convolutional Neural Networks for Land Cover Classification

Convolutional Neural Networks (CNN) have become the core of modern machine learning approaches. In addition to its inspiring interior design idea, the success of CNN depends mainly on two factors, the first is the availability of training data and the second is the computing power of the used devices. In the field of remote sensing, data availability is difficult and expensive. Furthermore, processing large remote sensing data to accommodate different models is a laborious process. At the same time, training data is often collected in the form of points distributed over crop fields rather than regions, which results in the scarcity of training data. To specifically address the scarcity of training points, in this paper we present a sparse pixel-based training of U-Net convolutional neural networks for land cover classification. Training images are reconstructed from the points’ collection in a random manner, they are used as an input for the convolution networks. Based on this proposed method, the amount of training data is reproduced from the different spectral signals for each land cover. We conducted extensive experiments on eight classes, using ground truth data collected from several locations in Fayoum Governorate, Egypt. The obtained results showed the superiority of the proposed method over other methods.

Collaborative Attention-Based Heterogeneous Gated Fusion Network for Land Cover Classification

Existing land cover classification methods mostly rely on either the optical or synthetic aperture radar (SAR) features alone, which ignore the mutual complementary effects between optical and SAR sources. In this article, we compare the distribution histograms of deep semantic features extracted from optical and SAR modalities within land cover categories, which intuitively demonstrates that there are the large complementary potentials between the optical and SAR features. Therefore, we propose a novel collaborative attention-based heterogeneous gated fusion network (CHGFNet), which hierarchically fuses both optical and SAR features for land cover classification. More specifically, the CHGFNet consists of three main components: two-stream feature extractor, multimodal collaborative attention module (MCAM), and the gated heterogeneous fusion module (GHFM). Given optical and SAR patch pairs, two-stream feature extractor introduces multistage feature learning methodology to acquire discriminative optical and SAR features. Then, to explore the inherent complementarity between optical and SAR features, MCAM is embedded into CHGFNet, which provides an efficient stage to capture the correlation between optical and SAR features by jointly calculating the collaborative attention in joint feature space. Finally, to automatically learn the varying contributions of both optical and SAR features for classifying different land categories, GHFM is used to fuse both optical and SAR features. Extensive comparative evaluations demonstrate the advantages of CHGFNet within land cover classification over the state-of-the-art methods on three co-registered optical and SAR data sets.

Fully convolutional networks for land cover classification from historical panchromatic aerial photographs

Historical aerial photographs provide salient information on the historical state of the landscape. The exploitation of these archives is often limited by accessibility and the time-consuming process of digitizing the analogue copies at a high resolution and processing them with a proper photogrammetric workflow. Furthermore, these data are characterised by limited spectral information since it occurs very often in a single band. Our work presents a first application of deep learning for the extraction of land cover from historical aerial panchromatic photographs of the African cities of Goma, Bukavu and Bujumbura. We evaluate the suitability of deep learning for land cover generation from a challenging dataset of photographs from the 1940s and 1950s that covers large geographical extents and is characterised by radiometric variations between dates and locations. A fully convolutional approach is investigated by considering two network architectures with different strategies of exploiting contextual information: one used atrous convolutional layers without downsampling, whereas the second network has both downsampling and learned upsampling convolutional layers (U-NET). The networks are trained to detect three main classes namely, buildings, high vegetation and a mixed class of bare land and low vegetation class. High overall accuracies of >90% in Goma-Gisenyi and Bukavu, and >85% in Bujumbura are obtained. This work provides a novel methodology that outperforms a baseline standard machine learning classifier for the exploitation of the vast archives of historical aerial photographs that can aid long-term environmental baseline studies. Future work will entail developing domain adaptation strategies in order to make the trained network robust for different image mosaics.

Comparing Fully Deep Convolutional Neural Networks for Land Cover Classification with High-Spatial-Resolution Gaofen-2 Images

Land cover is an important variable of the terrestrial ecosystem that provides information for natural resources management, urban sprawl detection, and environment research. To classify land cover with high-spatial-resolution multispectral remote sensing imagery is a difficult problem due to heterogeneous spectral values of the same object on the ground. Fully convolutional networks (FCNs) are a state-of-the-art method that has been increasingly used in image segmentation and classification. However, a systematic quantitative comparison of FCNs on high-spatial-multispectral remote imagery was not yet performed. In this paper, we adopted the three FCNs (FCN-8s, Segnet, and Unet) for Gaofen-2 (GF2) satellite imagery classification. Two scenes of GF2 with a total of 3329 polygon samples were used in the study area and a systematic quantitative comparison of FCNs was conducted with red, green, blue (RGB) and RGB+near infrared (NIR) inputs for GF2 satellite imagery. The results showed that: (1) The FCN methods perform well in land cover classification with GF2 imagery, and yet, different FCNs architectures exhibited different results in mapping accuracy. The FCN-8s model performed best among the Segnet and Unet architectures due to the multiscale feature channels in the upsampling stage. Averaged across the models, the overall accuracy (OA) and Kappa coefficient (Kappa) were 5% and 0.06 higher, respectively, in FCN-8s when compared with the other two models. (2) High-spatial-resolution remote sensing imagery with RGB+NIR bands performed better than RGB input at mapping land cover, and yet the advantage was limited; the OA and Kappa only increased an average of 0.4% and 0.01 in the RGB+NIR bands. (3) The GF2 imagery provided an encouraging result in estimating land cover based on the FCN-8s method, which can be exploited for large-scale land cover mapping in the future.

Land Cover Classification using Very High Spatial Resolution Remote Sensing Data and Deep Learning

Abstract The paper describes the process of training a convolutional neural network to segment land into its labelled land cover types such as grass, water, forest and buildings. This segmentation can promote automated updating of topographical maps since doing this manually is a time-consuming process, which is prone to human error. The aim of the study is to evaluate the application of U-net convolutional neural network for land cover classification using countrywide aerial data. U-net neural network architecture has initially been developed for use in biomedical image segmentation and it is one of the most widely used CNN architectures for image segmentation. Training data have been prepared using colour infrared images of Ventspils town and its digital surface model (DSM). Forest, buildings, water, roads and other land plots have been selected as classes, into which the image has been segmented. As a result, images have been segmented with an overall accuracy of 82.9 % with especially high average accuracy for the forest and water classes.

Dense Dilated Convolutions’ Merging Network for Land Cover Classification

Land cover classification of remote sensing images is a challenging task due to limited amounts of annotated data, highly imbalanced classes, frequent incorrect pixel-level annotations, and an inherent complexity in the semantic segmentation task. In this article, we propose a novel architecture called the dense dilated convolutions' merging network (DDCM-Net) to address this task. The proposed DDCM-Net consists of dense dilated image convolutions merged with varying dilation rates. This effectively utilizes rich combinations of dilated convolutions that enlarge the network's receptive fields with fewer parameters and features compared with the state-of-the-art approaches in the remote sensing domain. Importantly, DDCM-Net obtains fused local- and global-context information, in effect incorporating surrounding discriminative capability for multiscale and complex-shaped objects with similar color and textures in very high-resolution aerial imagery. We demonstrate the effectiveness, robustness, and flexibility of the proposed DDCM-Net on the publicly available ISPRS Potsdam and Vaihingen data sets, as well as the DeepGlobe land cover data set. Our single model, trained on three-band Potsdam and Vaihingen data sets, achieves better accuracy in terms of both mean intersection over union (mIoU) and F1-score compared with other published models trained with more than three-band data. We further validate our model on the DeepGlobe data set, achieving state-of-the-art result 56.2% mIoU with much fewer parameters and at a lower computational cost compared with related recent work. Code available at https://github.com/samleoqh/DDCM-Semantic-Segmentation-PyTorch

A multi-level context-guided classification method with object-based convolutional neural network for land cover classification using very high resolution remote sensing images

Classification of very high resolution imagery (VHRI) is challenging due to the difficulty in mining complex spatial and spectral patterns from rich image details. Various object-based Convolutional Neural Networks (OCNN) for VHRI classification have been proposed to overcome the drawbacks of the redundant pixel-wise CNNs, owing to their low computational cost and fine contour-preserving. However, classification performance of OCNN is still limited by geometric distortions, insufficient feature representation, and lack of contextual guidance. In this paper, an innovative multi-level context-guided classification method with the OCNN (MLCG-OCNN) is proposed. A feature-fusing OCNN, including the object contour-preserving mask strategy with the supplement of object deformation coefficient, is developed for accurate object discrimination by learning simultaneously high-level features from independent spectral patterns, geometric characteristics, and object-level contextual information. Then pixel-level contextual guidance is used to further improve the per-object classification results. The MLCG-OCNN method is intentionally tested on two validated small image datasets with limited training samples, to assess the performance in applications of land cover classification where a trade-off between time-consumption of sample training and overall accuracy needs to be found, as it is very common in the practice. Compared with traditional benchmark methods including the patch-based per-pixel CNN (PBPP), the patch-based per-object CNN (PBPO), the pixel-wise CNN with object segmentation refinement (PO), semantic segmentation U-Net (U-NET), and DeepLabV3+(DLV3+), MLCG-OCNN method achieves remarkable classification performance (> 80 %). Compared with the state-of-the-art architecture DeepLabV3+, the MLCG-OCNN method demonstrates high computational efficiency for VHRI classification (4–5 times faster).

A Hybrid Capsule Network for Land Cover Classification Using Multispectral LiDAR Data

Land cover mapping is an effective way to quantify land resources and monitor their changes. It plays an important role in a wide range of applications. This letter proposes a hybrid capsule network for land cover classification using multispectral light detection and ranging (LiDAR) data. First, the multispectral LiDAR data were rasterized into a set of feature images to exploit the geometrical and spectral properties of different types of land covers. Then, a hybrid capsule network composed of an encoder network and a decoder network is trained to extract both high-level local and global entity-oriented capsule features for accurate land cover classification. Quantitative classification evaluations on two data sets show that the overall accuracy, average accuracy, and kappa coefficient of over 97.89%, 94.54%, and 0.9713, respectively, are obtained. Comparative studies with five existing methods confirm that the proposed method performs robustly and accurately in land cover classification using the multispectral LiDAR data.

Fusion of Deep Convolutional Neural Networks for Land Cover Classification of High-Resolution Imagery

Deep convolutional neural networks (DCNNs) have recently emerged as a dominant paradigm for machine learning in a variety of domains. However, acquiring a suitably large data set for training DCNN is often a significant challenge. This is a major issue in the remote sensing domain, where we have extremely large collections of satellite and aerial imagery, but lack the rich label information that is often readily available for other image modalities. In this letter, we investigate the use of DCNN for land–cover classification in high-resolution remote sensing imagery. To overcome the lack of massive labeled remote-sensing image data sets, we employ two techniques in conjunction with DCNN: transfer learning (TL) with fine-tuning and data augmentation tailored specifically for remote sensing imagery. TL allows one to bootstrap a DCNN while preserving the deep visual feature extraction learned over an image corpus from a different image domain. Data augmentation exploits various aspects of remote sensing imagery to dramatically expand small training image data sets and improve DCNN robustness for remote sensing image data. Here, we apply these techniques to the well-known UC Merced data set to achieve the land–cover classification accuracies of 97.8 ± 2.3%, 97.6 ± 2.6%, and 98.5 ± 1.4% with CaffeNet, GoogLeNet, and ResNet, respectively.

Application of a Bayesian network for land-cover classification from a Landsat 7 ETM+ image

This article describes the use of a Bayesian network (BN) for the classification of land cover from satellite imagery in northern Swaziland. The main objective of this work was to apply and evaluate the efficacy of a BN for land-cover classification using gap-filled and terrain-corrected Landsat 7 Enhanced Thematic Mapper Plus (ETM+) imagery acquired on 15 May 2007. The posterior probabilities (parameters) were estimated using the expectation-maximization (EM) and conjugate gradient descent (CGD) algorithms. A comparison of the results obtained from the algorithms indicates similar and excellent overall classification accuracies of 93.01%, and kappa coefficient values of 0.9143. The main result obtained in this study is that both algorithms considered here provide relatively similar and accurate solutions for the classification of the multispectral image although the EM algorithm is marginally competitive relative to CGD algorithm when measured in terms of the Brier score and the logarithmic loss.

N-Version Programming of Artificial Neural Networks for Land Cover Classification from Satellite Data

N-Version Programming of Artificial Neural Networks for Land Cover Classification from Satellite Data

Reusing back-propagation artificial neural networks for land cover classification in tropical savannahs

This paper focuses on a method to overcome some of the disadvantages that are related with the use of artificial neural networks (ANNs) as supervised classifiers. The proposed method aims at speeding up network learning, improving classification accuracies and reducing variability on classification performance due to random weight initialization. This can be realized by transferring implicit knowledge from a previously learned source task to a new target task using the proposed algorithm, Discriminality Based Transfer (DBT). The presented approach is compared with conventional network training and a literal transfer method in a 13-class tropical savannah classification experiment using Landsat Thematic Mapper (TM) data. Knowledge was extracted from a network trained on the Kara experimental site in Togo. This information was used to classify the Savanes-L'Oti area which differs in terms of geographical position, image acquisition date, climatological condition and land cover. It was possible to speed up network learning 5.2, 4.3 and 1.8 times using, respectively, 5-, 10- and 20-pixels-per-class training sets. Larger training sets showed less speed improvement. After applying DBT, average classification accuracies were not significantly different from accuracies obtained after training random initialized networks, although DBT tended to show better performance on smaller training sets. It was possible to explain differences in individual class accuracies by analysing Bhattacharyya (BH) distances calculated between all Kara and Savanes-L'Oti classes. Finally, variability on classification performance decreased significantly when training with 5-, 10- and 20-pixels-per-class training sets after DBT application.