The accuracy of Digital Surface Models (DSMs) generated using stereo matching methods varies due to the varying acquisition conditions and configuration parameters of stereo images. It has been a good practice to fuse these DSMs generated from various stereo pairs to achieve enhanced, in which multiple DSMs are combined through computational approaches into a single, more accurate, and complete DSM. However, accurately characterizing detailed objects and their boundaries still present a challenge since most boundary-ware fusion methods still struggle to achieve sharpened depth discontinuities due to the averaging effects of different DSMs. Therefore, we propose a simple and efficient adaptive image-guided DSM fusion method that applies k-means clustering on small patches of the orthophoto to guide the pixel-level fusion adapted to the most consistent and relevant elevation points. The experiment results show that our proposed method has outperformed comparing methods in accuracy and the ability to preserve sharpened depth edges.

Accurate boundaries of smallholder farm fields are important and indispensable geo-information that benefits farmers, managers, and policymakers in terms of better managing and utilizing their agricultural resources. Due to their small size, irregular shape, and the use of mixed-cropping techniques, the farm fields of smallholder can be difficult to delineate automatically. In recent years, numerous studies on field contour extraction using a deep Convolutional Neural Network (CNN) have been proposed. However, there is a relative shortage of labeled data for filed boundaries, thus affecting the training effect of CNN. Traditional methods mostly use image flipping, and random rotation for data augmentation. In this paper, we propose to apply Generative Adversarial Network (GAN) for the data augmentation of farm fields label to increase the diversity of samples. Specifically, we propose an automated method featured by Fully Convolutional Neural networks (FCN) in combination with GAN to improve the delineation accuracy of smallholder farms from Very High Resolution (VHR) images. We first investigate four State-Of-The-Art (SOTA) FCN architectures, i.e., U-Net, PSPNet, SegNet and OCRNet, to find the optimal architecture in the contour detection task of smallholder farm fields. Second, we apply the identified optimal FCN architecture in combination with Contour GAN and pixel2pixel GAN to improve the accuracy of contour detection. We test our method on the study area in the Sudano-Sahelian savanna region of northern Nigeria. The best combination achieved F1 scores of 0.686 on Test Set 1 (TS1), 0.684 on Test Set 2 (TS2), and 0.691 on Test Set 3 (TS3). Results indicate that our architecture adapts to a variety of advanced networks and proves its effectiveness in this task. The conceptual, theoretical, and experimental knowledge from this study is expected to seed many GAN-based farm delineation methods in the future.

Structure-from-Motion (SfM) techniques have been widely used for 3D geometry reconstruction from multi-view images. Nevertheless, the efficiency and quality of the reconstructed geometry depends on multiple factors, i.e., the base-height ratio, intersection angle, overlap, and ground control points, etc., which are rarely quantified in real-world applications. To answer this question, in this paper, we take a data-driven approach by analyzing hundreds of terrestrial stereo image configurations through a typical SfM algorithm. Two main meta-parameters with respect to base-height ratio and intersection angle are analyzed. Following the results, we propose a Skeletal Camera Network (SCN) and embed it into the SfM to lead to a novel SfM scheme called SCN-SfM, which limits tie-point matching to the remaining connected image pairs in SCN. The proposed method was applied in three terrestrial datasets. Experimental results have demonstrated the effectiveness of the proposed SCN-SfM to achieve 3D geometry with higher accuracy and fast time efficiency compared to the typical SfM method, whereas the completeness of the geometry is comparable.

Bridge deformation monitoring usually adopts contact sensors, and the implementation process is often limited by the environment and observation conditions, resulting in unsatisfactory monitoring accuracy and effect. Ground-Based Synthetic Aperture Radar (GBSAR) combined with corner reflectors was used to perform static load-loaded deformation destruction experiments on solid model bridges in a non-contact manner. The semi parametric spline filtering and its optimization method were used to obtain the monitoring results of the GBSAR radar’s line of sight deformation, and the relative position of the corner reflector and the millimeter level deformation signals under different loading conditions were successfully extracted. The deformation transformation model from the radar line of sight direction to the vertical vibration direction was deduced. The transformation results of deformation monitoring and the measurement data such as the dial indicator were compared and analyzed. The occurrence and development process of bridge deformation and failure were successfully monitored, and the deformation characteristics of the bridge from continuous loading to eccentric loading until bridge failure were obtained. The experimental results show that GBSAR combined with corner reflector can be used for deformation feature acquisition, damage identification and health monitoring of bridges and other structures, and can provide a useful reference for design, construction and safety evaluation.

Timely and accurate population statistic data plays an important role in many fields. To illustrate the demographic characteristics, population density is a crucial factor in evaluating population data. With a dynamic regional migration in population, it is a challenging job to evaluate population density without a census-based survey. We present the approach to classify satellite images in different magnitudes in population density and execute the comparative experiment to discuss the factors that influence the identification to the images with the deep learning approach. In this paper, we use satellite imagery and community population density data. With convolutional neural networks, we evaluated the performance of CNN on population estimation with satellite images, found the features that are important in population estimation, and then perform the sensitive analysis.

Panoramic images are widely used in many scenes, especially in virtual reality and street view capture. However, they are new for street furniture identification which is usually based on mobile laser scanning point cloud data or conventional 2D images. This study proposes to perform semantic segmentation on panoramic images and transformed images to separate light poles and traffic signs from background implemented by pre-trained Fully Convolutional Networks (FCN). FCN is the most important model for deep learning applied on semantic segmentation for its end to end training process and pixel-wise prediction. In this study, we use FCN-8s model that pre-trained on cityscape dataset and finetune it by our own data. Then replace cross entropy loss function with focal loss function in the FCN model and train it again to produce the predictions. The results show that in all results from pre-trained model, fine-tuning, and FCN model with focal loss, the light poles and traffic signs are detected well and the transformed images have better performance than panoramic images in the prediction according to the Recall and IoU evaluation.

In this paper, an efficient technique for removing strip noise from remote sensing images is proposed in order to better retain image details. Firstly, the remote sensing image with strip noise is decomposed by wavelet technology; Secondly, two variational models are constructed, stripe preserve variation model and a destriping variation model. In order to efficiently separate the detail information in the low level high-frequency component, the stripe preserve variation model eliminates the detail information from the low level high-frequency component (including strip noise) while maintaining the strip noise (including strip noise). In order to successfully save the details in the high level high-frequency component, the destriping variation model eliminates the strip noise in the high level high-frequency component (including the strip noise). Finally, wavelet reconstruction is used to get the denoised image. It is clear from a comparison with previous approaches that the suggested method not only successfully removes strip noise but also preserves image details.