Evaluation of U-Net + + architectures in high resolution image classification for urban planning

Conventional geographic object-based image analysis (GEOBIA) land cover classification methods by using very high resolution images are hardly applicable due to their complex ground truth and manually selected features, while convolutional neural networks (CNNs) with many hidden layers provide the possibility of extracting deep features from very high resolution images. Compared with pixel-based CNNs, superpixel-based CNN classification, carrying on the idea of GEOBIA, is more efficient. However, superpixel-based CNNs are still problematic in terms of their processing units and accuracies. Firstly, the limitations of salt and pepper errors and low boundary adherence caused by superpixel segmentation still exist; secondly, this method uses the central point of the superpixel as the classification benchmark in identifying the category of the superpixel, which does not allow classification accuracy to be ensured. To solve such problems, this paper proposes a region-based majority voting CNN which combines the idea of GEOBIA and the deep learning technique. Firstly, training data was manually labeled and trained; secondly, images were segmented under multiresolution and the segmented regions were taken as basic processing units; then, point voters were generated within each segmented region and the perceptive fields of points voters were put into the multi-scale CNN to determine their categories. Eventually, the final category of each region was determined in the region majority voting system. The experiments and analyses indicate the following: 1. region-based majority voting CNNs can fully utilize their exclusive nature to extract abstract deep features from images; 2. compared with the pixel-based CNN and superpixel-based CNN, the region-based majority voting CNN is not only efficient but also capable of keeping better segmentation accuracy and boundary fit; 3. to a certain extent, region-based majority voting CNNs reduce the impact of the scale effect upon large objects; and 4. multi-scales containing small scales are more applicable for very high resolution image classification than the single scale.

We propose a non-homogeneous Conditional Random Field built over an adjacency graph of superpixels for contextual classification of high-resolution satellite images. By introducing the contextual histogram descriptor, our model includes spatially dependent unary and pairwise potentials that capture contextual interactions of the data as well as the labels. This results the non-homogeneity of the fields which improves the accuracy of the classification. Furthermore, our discriminative model performs a multi-cue combination by incorporating efficiently color, texture, edge, curvilinear continuity and familiar configuration cues. As for potentials, both local and global feature functions are learned using joint boosting whereas a likelihood ratio is learned to derive the pairwise edge potential. In this model, the optimal scene interpretation is inferred using a cluster sampling method, the Swendsen-Wang Cut algorithm. Promising results are shown on SPOT-5 satellite images.

The classification of image is essential to make LULC (Land Use Land Cover) maps. However, the classification of land cover plays a vital role for studying and modernizing the land areas. Recently, deep learning (DL) techniques have achieved outstanding performance in the classification of high-resolution images. Different techniques have been employed in traditional methods to identify LULC due to its complex and ever-changing nature. However, these studies have shown improved outcomes despite some restrictions such as inaccuracies and reduced performance. To address these problems, the proposed study introduces a Squeeze Synchronization Layer (SSL) and a Convolve Craft Focus Module (CCFM) where, SSL reduces input data complexity by removing noise and irrelevant information from images using pooling and convolutional operations also, CCFM enhances feature extraction to improve land classification accuracy. The EUROSAT land image dataset is utilized for the evaluation of the introduced model. Whereas, the dataset comprises of 64x64 images, which are captured by satellite Sentinel-2A in ResNet 101 input layer. Although, a SSL is suggested, and a CCFM is implemented in the convolutional layer for classifying land images. However, the efficiency of the system is evaluated by measuring performance metrics such as recall, F1-score, precision, and accuracy values of the proposed system. The accuracy value of the proposed system is 96% of accuracy, 100% of precision, 100% of recall, and 100% of F1-score, signifies the superior efficiency of the proposed model

Super-resolution guided knowledge distillation for low-resolution image classification

Remotely sensed image grading advanced considerably; takes into account the availability and abundance of various resolution image grading algorithms. Several works were successful by the fusion of space-spectrum knowledge with supporting vector machines (SVM). To incorporate all these data with the composite approach, we suggest a technique using a hybrid multi-spectral and multi-instance procedure. This paper introduces a groundbreaking approach to exploring urban buildings through the implementation of the SVM-based support classification and Multi-instance learning (MIL). In this paper, we present the use of this model, the classification of images. Use high-resolution technology from Quickbird. This practice and archery have contributed to the performance, efficiency, and power. The suggested solution was tested in traditional urban imagery scenes. The results show a major improvement the classification performance compared to the two separately used attributes. The results of the experiments indicate a very promising accuracy of 91, 24%.

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.

An object-oriented, multiscale feature extraction approach is proposed for the land-cover classification of high spatial resolution images. The approach provides more discriminative features by considering the spatial context information from different segmentation levels. It consists of three successive substeps: segmentation by mean-shift algorithm, an iteratively merging process controlled by merging cost function and range-of-scale parameter, and feature extraction from linked multilevel image partitions. The mean-shift method is to get boundary-preserved and spectrally homogeneous over-segmentation regions. Then, a family of nested image partitions is constructed by a merging procedure. Meanwhile, every region of the finest scale is linked to image objects of its superlevels. Finally, every region in the finest scale is treated as a basic analysis unit, and the feature vectors are created by stacking statistics from the region and their superlevels. A support vector machine is used as a classifier and the method on two widely used high spatial resolution data sets over Pavia City, Italy, are evaluated. Compared with results reported in many papers, the result indicates superior accuracy.

Markov random field (MRF) based methods have been widely used in high spatial resolution (HSR) image classification. However, many existing MRF-based methods put more emphasis on pixel level contexts while less on superpixel level contextual information. To cope with this issue, this article presents a novel bilevel contextual MRF framework, named BLC-MRF, for HSR imagery classification. Specifically, pixel and superpixel level dependence are incorporated into the proposed MRF model to fully exploit spectral–spatial contextual information and preserve object boundaries in HSR images. In BLC-MRF, a pixel level MRF model is first performed and then cascaded as an input of a superpixel level MRF. In superpixel level, unary and pairwise potential terms are constructed by using the superpixel probability estimation method and spectral histogram distance, respectively. At last, a contextual MRF model is conducted and the final classification map can be computed by using $\alpha$ -expansion algorithm. The benefits of BLC-MRF are twofold: first, the pixel and superpixel level contextual information can be exploited under MRF framework to preserve object boundaries for improving the classification performance, and, second, the algorithm can provide promising results with a small number of training samples. Experimental results on three HSR datasets demonstrate that the proposed approach outperforms several state-of-the-art methods in terms of the classification performance.

This paper presents a genetic algorithm (GA) approach for parameters optimization of support vector machine, which is used for the object-oriented classification of high spatial resolution images over urban area. The proposed method is a three-step routine involves the integration of 1) image segmentation, 2) GA-based parameter optimization of Support vector machine (SVM), and 3) objected-based classification. Experiments conducted on multi-spectral Quick-Bird image fused with panchromatic image in Fuzhou city. In addition, a traditional parameter searching method, Grid algorithm, was investigated to evaluate the effectiveness of the proposed approach. The results show that our proposed GA-based approach significantly outperforms the Grid algorithm both in terms of classification accuracy and time efficiency.

Classification of high resolution urban remote sensing imagery is addressed. The classification is done by both considering the panchromatic imagery and the multi-spectral image obtained using the spectrally consistent fusion method introduced in [1]. The data are classified using support vector machines (SVM). To further enhance the classification accuracy, mathematical morphology is used to derive local spatial information from the panchromatic data. In particular we use the Morphological Profile (MP) in classification of satellite imagery as was proposed in [2, 3]. We also use the derivative of the MP (DMP). In the majority of the image fusion (pansharpening) techniques proposed today, there is a compromise between the spatial enhancement and the spectral consistency. By comparing classification results obtained by using our model based scheme [1] to results obtained using the IHS and Brovey fusion methods, we find that spectrally consistent data give better results when it comes to classification.

Convolutional neural networks (CNNs) have been applied to learn spatial features for high-resolution (HR) synthetic aperture radar (SAR) image classification. However, there has been little work on integrating the unique statistical distributions of SAR images which can reveal physical properties of terrain objects, into CNNs in a supervised feature learning framework. To address this problem, a novel end-to-end supervised classification method is proposed for HR SAR images by considering both spatial context and statistical features. First, to extract more effective spatial features from SAR images, a new deep spatial context encoder network (DSCEN) is proposed, which is a lightweight structure and can be effectively trained with a small number of samples. Meanwhile, to enhance the diversity of statistics, the nonstationary joint statistical model (NS-JSM) is adopted to form the global statistical features. Specifically, SAR images are transformed into the Gabor wavelet domain and the produced multi-subbands magnitudes and phases are modeled by the log-normal and uniform distribution. The covariance matrix is further utilized to capture the inter-scale and intra-scale nonstationary correlation between the statistical subbands and make the joint statistical features more compact and distinguishable. Considering complementary advantages, a feature fusion network (Fusion-Net) base on group compression and smooth normalization is constructed to embed the statistical features into the spatial features and optimize the fusion feature representation. As a result, our model can learn the discriminative features and improve the final classification performance. Experiments on four HR SAR images validate the superiority of the proposed method over other related algorithms.

We proposed a new method based on supervised locality preserving projections (SLPP) for classification of high resolution satellite images. Compared with other subspace methods such as PCA and ICA, SLPP can preserve local geometric structure of data and enhance within-class local information. The proposed method has been successfully applied to IKONOS images and experimental results show that the proposed SLPP based method outperform ICA-based method. The proposed method can be practically incorporated into a GIS system.KeywordsHigh-resolution satellite imagesupervisedlocality preserving projectionsclassificationmulti-spectral image

In this paper, we propose an approach to the classification of high-resolution hyperspectral images in the applied problem of identification of vegetation types. A modified spectral-spatial convolutional neural network with compensation for illumination variations is used as a classifier. For generating a training dataset, an algorithm based on an adaptive vegetation index is proposed. The effectiveness of the proposed approach is shown on the basis of survey data of agricultural lands obtained from a compact hyperspectral camera developed in-house.

The paper proposes approaches for the classification of high-resolution hyperspectral images in the problem of classification of soil species classification. A spectral-spatial convolutional neural network with compensation for illumination variations is used as a classifier. The effectiveness of the proposed approach in the problem of classification of hyperspectral images of soils obtained by a scanning type hyperspectrometer is shown. A multiclass neural network is compared with an ensemble in which the results of a multiclass neural network are refined by several binary classifiers. It is shown that the use of normalization of illumination inhomogeneity and the use of an ensemble of convolutional spatial-spectral neural networks can significantly increase the accuracy of soil type classification.

The segmentation and classification of high-resolution satellite images (HRSI) are useful approaches to extract information. In recent times, roads and buildings have been classified for analysis of urban areas in a better manner. Apart from these, healthy trees are also an important factor in HRSI, i.e. adjacent to roads, and vegetation. They reflect the area in an image as land cover. Other important information, shadow, is extracted from satellite images, which indicates the presence of trees and built-up areas such as buildings, flyovers, etc. In this article, a weighted membership-function-based fuzzy c-means with spatial constraints (WMFCSC) approach for automated satellite image classification is proposed. Initially, spatially fuzzy clustering is used to classify the satellite images in healthy trees with vegetation, roads, and shadows, which includes the information of spatial constraints. The road results of the classified image are still having non-road segments. Therefore, the proposed four intermediate stages (IS) are used to extract the road information, followed by the results of road areas of the WMFCSC approach. The framework of IS helps to remove the false road segments which are adjacent to roads and renovates the segmented roads due to the shadow effect. A final step of a hybrid WMFCSC-IS approach is used to extract the road network. The results of classified images confirm the effectiveness of the WMFCSC-IS approach for satellite image classification.