Scalable Machine Learning Approaches for Neighborhood Classification Using Very High Resolution Remote Sensing Imagery

The detection of building edges from very high resolution (VHR) remote sensing imagery is essential to various geo-related applications, including surveying and mapping, urban management, etc. Recently, the rapid development of deep convolutional neural networks (DCNNs) has achieved remarkable progress in edge detection; however, there has always been the problem of edge thickness due to the large receptive field of DCNNs. In this paper, we proposed a multi-scale erosion network (ME-Net) for building edge detection to crisp the building edge through two innovative approaches: (1) embedding an erosion module (EM) in the network to crisp the edge and (2) adding the Dice coefficient and local cross entropy of edge neighbors into the loss function to increase its sensitivity to the receptive field. In addition, a new metric, Ene, to measure the crispness of the predicted building edge was proposed. The experiment results show that ME-Net not only detects the clearest and crispest building edges, but also achieves the best OA of 98.75%, 95.00% and 95.51% on three building edge datasets, and exceeds other edge detection networks 3.17% and 0.44% at least in strict F1-score and Ene. In a word, the proposed ME-Net is an effective and practical approach for detecting crisp building edges from VHR remote sensing imagery.

Offshore and inland river ship detection has been studied on both synthetic aperture radar (SAR) and optical remote sensing imagery. However, the classic ship detection methods based on SAR images can cause a high false alarm ratio and be influenced by the sea surface model, especially on inland rivers and in offshore areas. The classic detection methods based on optical images do not perform well on small and gathering ships. This paper adopts the idea of deep networks and presents a fast regional-based convolutional neural network (R-CNN) method to detect ships from high-resolution remote sensing imagery. First, we choose GaoFen-2 optical remote sensing images with a resolution of 1 m and preprocess the images with a support vector machine (SVM) to divide the large detection area into small regions of interest (ROI) that may contain ships. Then, we apply ship detection algorithms based on a region-based convolutional neural network (R-CNN) on ROI images. To improve the detection result of small and gathering ships, we adopt an effective target detection framework, Faster-RCNN, and improve the structure of its original convolutional neural network (CNN), VGG16, by using multiresolution convolutional features and performing ROI pooling on a larger feature map in a region proposal network (RPN). Finally, we compare the most effective classic ship detection method, the deformable part model (DPM), another two widely used target detection frameworks, the single shot multibox detector (SSD) and YOLOv2, the original VGG16-based Faster-RCNN, and our improved Faster-RCNN. Experimental results show that our improved Faster-RCNN method achieves a higher recall and accuracy for small ships and gathering ships. Therefore, it provides a very effective method for offshore and inland river ship detection based on high-resolution remote sensing imagery.

Greenhouses are densely distributed across the cultivated land in high-resolution remote sensing imagery, resulting in the problem of dense object extraction. On the one hand, objects tend to be wrongly merged into one object since objects are connected; on the other hand, the existed random sampling scheme does not make good use of the object distribution density to improve the training effect. To meet the demand for dense greenhouse extraction, this paper proposes a novel deep learning-based greenhouse extraction algorithm. To solve the problem that dense objects tend to be wrongly merged, this paper proposes a dual-task learning module, which uses the discriminant object boundary of the pixel-based branch to distinguish the objects of the object-based branch; to take advantage of the object distribution density for effective training, a high-density biased sampler is proposed. Moreover, this paper provides a dataset of manually labeled imagery to train and develop the proposed method. Results on greenhouse extraction in six regions in China achieve peak mIoU and mAP value, surpassing the state-of-the-art methods. Finally, a product of a greenhouse map in China is provided for analysis.

Building extraction based on high-resolution remote sensing imagery has been widely used in automatic surveying and mapping. Recently, the instance segmentation algorithm has been introduced to the building extraction, which can calculate the number and area of buildings simultaneously. However, there are some challenges: 1) multi-scale buildings; 2) occlusion by other adjacent buildings. In this paper, to solve these problems, we propose a multi-scale building instance extraction framework based on feature pyramid object-aware convolution neural network (CNN). In order to solve the multi-scale problem, a feature pyramid CNN is proposed, which combines features from both the bottom-up and top-down architectures. In order to solve the occlusion problem, a multi-scale object-aware instance proposal network is proposed, which introduces the multiscale attention mechanism to aware objects. The experiments conducted on two public datasets and a self-constructed dataset of Changzhou show that the proposed method can achieve an excellent performance.

A new algorithm based on the fusion of edge and multispectral information is proposed for the pixel-wise classification of very high-resolution (VHR) remotely sensed imagery. It integrates the multispectral, spatial and structural information existing in the image. The edge feature is first extracted using an improved multispectral edge detection method, which takes into account the original multispectral bands, the linear NDVI, and the independent spectral components extracted by independent component analysis (ICA). Direction-lines are then defined using the edge and multispectral information. Two effective spatial measures are calculated based on the direction-lines in order to describe the contextual information and strengthen the multispectral feature space. Then, the support vector machine (SVM) is employed to classify the hybrid structural-multispectral feature set. In experiments, the proposed spatial measures were compared with the pixel shape index (PSI) and the gray level co-occurrence matrix (GLCM). The experimental results show that the proposed algorithm performs well in terms of classification accuracies and visual interpretation.

With the emergence of large scale Very High Resolution (VHR) remotely sensed imageries, more accurate urban building segmentation now become possible. Meanwhile, various appearances along with complicated background of the urban VHR imageries make accurate segmentation of urban buildings yet a very challenging task. In this study, we proposed DeepAttentionUnet, which is an end-to-end attention-based model following the basic structure of U-Net and being integrated with attention mechanism so that the model can automatically learn to focus on target building structure of different sizes and shapes. This model adopts attention mechanism to implicitly highlight useful features of buildings while suppressing irrelevant areas of input images and combines with residual learning method to ease training process by alleviating the degradation problem in the training process that often occurs in such applications. Our proposed DeepAttentionUnet was tested with a set of 0.075m resolution aerial images and was compared for performance under the exact same conditions with other two state-of-the-art segmentation networks, i. e. SegNet and U-Net, respectively. The experiment results suggested that in both quantitative and visual evaluation, our proposed model outperformed other two models in urban building semantic segmentation tasks in terms of F1 score, Kappa coefficient and overall accuracy, moreover, it also takes the advantage of having much less parameters compared with other two state-of-art models in comparison.

A comparison of selected classification algorithms for mapping bamboo patches in lower Gangetic plains using very high resolution WorldView 2 imagery

Abstract. The Support Vector Machine (SVM) is a theoretically superior machine learning methodology with great results in classification of remotely sensed datasets. Determination of optimal parameters applied in SVM is still vague to some scientists. In this research, it is suggested to use the Taguchi method to optimize these parameters. The objective of this study was to detect tree crowns on very high resolution (VHR) aerial imagery in Zagros woodlands by SVM optimized by Taguchi method. A 30 ha plot of Persian oak (Quercus persica) coppice trees was selected in Zagros woodlands, Iran. The VHR aerial imagery of the plot with 0.06 m spatial resolution was obtained from National Geographic Organization (NGO), Iran, to extract the crowns of Persian oak trees in this study. The SVM parameters were optimized by Taguchi method and thereafter, the imagery was classified by the SVM with optimal parameters. The results showed that the Taguchi method is a very useful approach to optimize the combination of parameters of SVM. It was also concluded that the SVM method could detect the tree crowns with a KHAT coefficient of 0.961 which showed a great agreement with the observed samples and overall accuracy of 97.7% that showed the accuracy of the final map. Finally, the authors suggest applying this method to optimize the parameters of classification techniques like SVM.

This study evaluates the impact of three Feature Selection (FS) algorithms in an Object Based Image Analysis (OBIA) framework for Very-High-Resolution (VHR) Land Use-Land Cover (LULC) classification. The three selected FS algorithms, Correlation Based Selection (CFS), Mean Decrease in Accuracy (MDA) and Random Forest (RF) based Recursive Feature Elimination (RFE), were tested on Support Vector Machine (SVM), K-Nearest Neighbor, and Random Forest (RF) classifiers. The results demonstrate that the accuracy of SVM and KNN classifiers are the most sensitive to FS. The RF appeared to be more robust to high dimensionality, although a significant increase in accuracy was found by using the RFE method. In terms of classification accuracy, SVM performed the best using FS, followed by RF and KNN. Finally, only a small number of features is needed to achieve the highest performance using each classifier. This study emphasizes the benefits of rigorous FS for maximizing performance, as well as for minimizing model complexity and interpretation.

In this paper, we present and apply a semisupervised support vector machine based on cluster kernels for the problem of very high resolution image classification. In the proposed setting, a base kernel working with labeled samples only is deformed by a likelihood kernel encoding similarities between unlabeled examples. The resulting kernel is used to train a standard support vector machine (SVM) classifier. Experiments carried out on very high resolution (VHR) multispectral and hyperspectral images using very few labeled examples show the relevancy of the method in the context of urban image classification. Its simplicity and the small number of parameters involved make it versatile and workable by unexperimented users.

In very high resolution (VHR) remote sensing (RS) classification tasks, conventional pixel-based contextual information extraction methods such as morphological profiles (MPs), extended MPs (EMPs) and MPs with partial reconstruction (MPPR) with limited numbers, sizes and shapes of structural elements (SEs) cannot perfectly match all sizes and shapes of the objects in an image. To overcome such limitation, we introduce novel spatial feature extractors, namely, the superpixel-guided morphological profiles (SPMPs), where the superpixels are used as SEs in opening by reconstruction and closing by reconstruction operations. Moreover, to avoid possible side effects from unusual maximum and minimum values within superpixels, the mean pixel value of superpixels is adopted (SPMPsM). Additionally, new decision forest based on penalizing the attributes in previous trees, the ForestPA is introduced and evaluated through a comparative investigation on three VHR multi-/hyperspectral RS image classification tasks. Support vector machine and benchmark ensemble classifiers, including bagging, AdaBoost, MultiBoost, ExtraTrees, Random Forest and Rotation Forest, are adopted. The experimental results confirm the effectiveness and superior performances of the proposed SPMPs and SPMPsM relative to those of the MPs and MPPR. Moreover, ForestPA outperforms only bagging and is not suitable for learning from large numbers of samples with high dimensionality from the computational efficiency and classification accuracy perspective.

Building extraction currently is important in the application of high-resolution remote sensing imagery. At present, quite a few algorithms are available for detecting building information, however, most of them still have some obvious disadvantages, such as the ignorance of spectral information, the contradiction between extraction rate and extraction accuracy. The purpose of this research is to develop an effective method to detect building information for Chinese GF-1 data. Firstly, the image preprocessing technique is used to normalize the image and image enhancement is used to highlight the useful information in the image. Secondly, multi-spectral information is analyzed. Subsequently, an improved morphological building index (IMBI) based on remote sensing imagery is proposed to get the candidate building objects. Furthermore, in order to refine building objects and further remove false objects, the post-processing (e.g., the shape features, the vegetation index and the water index) is employed. To validate the effectiveness of the proposed algorithm, the omission errors (OE), commission errors (CE), the overall accuracy (OA) and Kappa are used at final. The proposed method can not only effectively use spectral information and other basic features, but also avoid extracting excessive interference details from high-resolution remote sensing images. Compared to the original MBI algorithm, the proposed method reduces the OE by 33.14% .At the same time, the Kappa increase by 16.09%. In experiments, IMBI achieved satisfactory results and outperformed other algorithms in terms of both accuracies and visual inspection

Extracting road features from high-resolution remote sensing imagery is crucial for urban planning, navigation systems. However, this task is challenged by factors such as occlusions from buildings, interruptions in road continuity, and variations in road width and appearance. These issues often lead to segmentation discontinuities and misclassifications. This paper presents the Urban Road Extraction Network (UBR-Net), an architecture that enhances the DeepLabv3+ model to address these challenges. The key contributions lie in the specific design and integration of several modules. We introduce Cross-Residual Encoding blocks designed to preserve fine-grained details and mitigate the gradient vanishing problem, thereby improving road continuity. Additionally, UBR-Net incorporates a Multiscale Context Features Extraction (MCFE) module, enhanced with an Improved Self-Attention Block (ISAB), to capture rich, hierarchical feature representations with a focus on long-range dependencies. A Channel Spatial Attention Module (CSAM) is also integrated to refine the feature extraction process by focusing on critical channels and spatial regions. Evaluations on public datasets, including DeepGlobe and Massachusetts, show that UBR-Net reduces extraction errors from occlusions, achieving a 75.18% F1 score and a 57.22% Intersection over Union (IoU), surpassing existing methods. These results highlight UBR-Net’s effectiveness for urban analysis and its potential for more precise and efficient urban planning.

Deep convolutional neural networks have promoted significant progress in building extraction from high-resolution remote sensing imagery. Although most of such work focuses on modifying existing image segmentation networks in computer vision, we propose a new network in this paper, Deep Encoding Network (DE-Net), that is designed for the very problem based on many lately introduced techniques in image segmentation. Four modules are used to construct DE-Net: the inception-style downsampling modules combining a striding convolution layer and a max-pooling layer, the encoding modules comprising six linear residual blocks with a scaled exponential linear unit (SELU) activation function, the compressing modules reducing the feature channels, and a densely upsampling module that enables the network to encode spatial information inside feature maps. Thus, DE-Net achieves state-of-the-art performance on the WHU Building Dataset in recall, F1-Score, and intersection over union (IoU) metrics without pre-training. It also outperformed several segmentation networks in our self-built Suzhou Satellite Building Dataset. The experimental results validate the effectiveness of DE-Net on building extraction from aerial imagery and satellite imagery. It also suggests that given enough training data, designing and training a network from scratch may excel fine-tuning models pre-trained on datasets unrelated to building extraction.

Despite the substantial impact of rivers on the global marine litter problem, riverine litter has been accorded inadequate consideration. Therefore, our objective was to detect riverine litter by utilizing middle-scale multispectral satellite images and machine learning (ML), with the Tisza River (Hungary) as a study area. The Very High Resolution (VHR) images obtained from the Google Earth database were employed to recognize some riverine litter spots (a blend of anthropogenic and natural substances). These litter spots served as the basis for training and validating five supervised machine-learning algorithms based on Sentinel-2 images [Artificial Neural Network (ANN), Support Vector Classifier (SVC), Random Forest (RF), Naïve Bays (NB) and Decision Tree (DT)]. To evaluate the generalization capability of the developed models, they were tested on larger unseen data under varying hydrological conditions and with different litter sizes. Besides the best-performing model was used to investigate the spatio-temporal variations of riverine litter in the Middel Tisza. According to the results, almost all the developed models showed favorable metrics based on the validation dataset (e.g., F1-score; SVC: 0.94, ANN: 0.93, RF: 0.91, DT: 0.90, and NB: 0.83); however, during the testing process, they showed medium (e.g., F1-score; RF:0.69, SVC: 0.62; ANN: 0.62) to poor performance (e.g., F1-score; NB: 0.48; DT: 0.45). The capability of all models to detect litter was bounded to the pixel size of the Sentinel-2 images. Based on the spatio-temporal investigation, hydraulic structures (e.g., Kisköre Dam) are the greatest litter accumulation spots. Although the highest transport rate of litter occurs during floods, the largest litter spot area upstream of the Kisköre Dam was observed at low stages in summer. This study represents a preliminary step in the automatic detection of riverine litter; therefore, additional research incorporating a larger dataset with more representative small litter spots, as well as finer spatial resolution images is necessary.