We propose a new convolutional neural networks method in combination with ordinal regression aiming at assessing the degree of building damage caused by earthquakes with aerial imagery. The ordinal regression model and a deep learning algorithm are incorporated to make full use of the information to improve the accuracy of the assessment. A new loss function was introduced in this paper to combine convolutional neural networks and ordinal regression. Assessing the level of damage to buildings can be considered as equivalent to predicting the ordered labels of buildings to be assessed. In the existing research, the problem has usually been simplified as a problem of pure classification to be further studied and discussed, which ignores the ordinal relationship between different levels of damage, resulting in a waste of information. Data accumulated throughout history are used to build network models for assessing the level of damage, and models for assessing levels of damage to buildings based on deep learning are described in detail, including model construction, implementation methods, and the selection of hyperparameters, and verification is conducted by experiments. When categorizing the damage to buildings into four types, we apply the method proposed in this paper to aerial images acquired from the 2014 Ludian earthquake and achieve an overall accuracy of 77.39%; when categorizing damage to buildings into two types, the overall accuracy of the model is 93.95%, exceeding such values in similar types of theories and methods.

The aim of this research is to develop a technique that can detect changes in urban buildings after natural disasters such as earthquakes, typhoons or tsunamis. The increasing demands for real-time and effective change detection methodologies for urban buildings, including collapsed, height-changed and removed buildings, require the use of high spatial resolution images. Conventional change detection techniques, developed for medium resolution remotely sensed data, are neither effective nor efficient due to their difficulties in discriminating real changes from changes caused by change in perspective view, as well as due to the increased spatial detail in texture and edges in high spatial resolution images. In this paper, an automatic change detection method, using UltraCamD images and existing 3-dimensional computer-aided design (CAD) data, which analyse changes at the building level, is presented. The edge features of every building in the image are first extracted, using the following steps: pre-processing, edge extraction using a Canny detector, edge denoising, and line segmentation. Then, the silhouettes of each building, which are extracted from the CAD model based on camera parameters, are projected onto the image and are compared to the edge features of each building in the image. The portion Hausdorff matching methodology is employed to calculate the similarities between the projected silhouettes of each building from the CAD model, and the edge features of each building are extracted from the image. Using the calculated similarity, collapsed or removed buildings can be detected. The maximal similarity between the edge features and the projected silhouettes in the direction of projection is used to detect changes in building height. The methods are illustrated with an airborne UltraCamD image over the city of Tokyo, Japan. The results indicate that among 1320 buildings (10 were collapsed) used in the test, 45 buildings were successfully detected to have had collapsed or been removed. Nine out of the forty-five buildings were correctly detected from the list of 10 collapsed buildings. Overall, 2885 buildings were used to test the method for detecting changes in building height, of which 11 of the 27 buildings known to have changed heights with the CAD model were successfully detected. However, our methodology faces challenges from factors such as CAD model errors, high building density, and small changes in building height.

Taking the Guanzhong Plain of China as the research area, a method combining the trend surface analysis (TSA) method and the spatial variation weight method (SVWM) was used to upscale the vegetation temperature condition index (VTCI) retrieved from Landsat 8 from a finer spatial resolution to a coarser. The upscaled results were compared with VTCIs retrieved from Aqua MODIS to provide technical support for the comprehensive application of drought monitoring results on two spatial scales. Meanwhile, the upscaled methods were systematically evaluated in a case study according to various indicators, including the spatial distribution characteristics for drought, the statistical characteristics, the semivariogram function (SVF), the structural similarity (SSIM), the correlation coefficient (r), and the root mean square error (RMSE). The results show that the dominant class variability weighted (DCVW) method performed better than did the arithmetic average variability weighted (AAVW) and median pixel variability weighted (MPVW) methods in terms of the statistical characteristics, SSIM, correlation and RMSE, and the upscaling model based on the union of TSA and DCVW had the highest accuracy. The DCVW is a promising tool for upscaling Landsat-VTCI images of the Guanzhong Plain from finer to coarser spatial resolutions because of its efficiency and flexibility.

Improving the speed and accuracy of earthquake disaster loss evaluations is very important for disaster response and rescue. This paper presents a new method for urban damage assessments after earthquake disasters by using a change detection technique between bi-temporal (pre- and post-event) high-resolution optical images. A similarity index derived from a pair of images was used to extract the characteristics of collapsed buildings. In this paper, the methods are illustrated using two case studies. Our results confirmed the effectiveness and precision of the proposed technique with optical data of the damage presented using a block scale.

An automatic method is developed here to paste aerial images onto an urban 3D surface model for more realistic visualization. In this study we extracted different side views of urban constructions from an aerial image acquired with an airborne linear scanner sensor. We then matched those sideview images onto a 3D surface model according to the correspondence between the image and model. Side view feature extraction from images and matching those features to 3D models are two key steps in developing an automatic 3D image modelling technique. Here we present a new line-extraction approach using a multiple-level feature filter, which consists of the following: a Canny edge detector, an edge phase filter, an edge direction filter with fault tolerance, a Hough transformer, and a neighbouring line-segment fuser. We propose a base-line segmentation and parallelogram extraction algorithm based on perceptual organization. The algorithm employs uncertainty reasoning and is based on part forms for shape expression. It is computationally less intensive and noise free. Matching 2D images to 3D models requires finding a transformation matrix to minimize error. A lot of algorithms have been presented to solve the matching problem. However, there is still no good solution to the problem as it has too many unknown parameters. In this research, we first project images based on the camera model after a partial matching between the extracted parallelogram and the 3D model is carried out. Then, the Hausdorff distance is calculated between edges in the original image and the projected image, based on which sideview feature mapping is realized to obtain 3D virtual views based on a 3D surface model and a 2D image.