StyHighNet: Semi-Supervised Learning Height Estimation from a Single Aerial Image via Unified Style Transferring

The Use of Deep Learning Methods for Object Height Estimation in High Resolution Satellite Images

Processing single high-resolution satellite images may provide a lot of important information about the urban landscape or other applications related to the inventory of high-altitude objects. Unfortunately, the direct extraction of specific features from single satellite scenes can be difficult. However, the appropriate use of advanced processing methods based on deep learning algorithms allows us to obtain valuable information from these images. The height of buildings, for example, may be determined based on the extraction of shadows from an image and taking into account other metadata, e.g., the sun elevation angle and satellite azimuth angle. Classic methods of processing satellite imagery based on thresholding or simple segmentation are not sufficient because, in most cases, satellite scenes are not spectrally heterogenous. Therefore, the use of classical shadow detection methods is difficult. The authors of this article explore the possibility of using high-resolution optical satellite data to develop a universal algorithm for a fully automated estimation of object heights within the land cover by calculating the length of the shadow of each founded object. Finally, a set of algorithms allowing for a fully automatic detection of objects and shadows from satellite and aerial imagery and an iterative analysis of the relationships between them to calculate the heights of typical objects (such as buildings) and atypical objects (such as wind turbines) is proposed. The city of Warsaw (Poland) was used as the test area. LiDAR data were adopted as the reference measurement. As a result of final analyses based on measurements from several hundred thousand objects, the global accuracy obtained was ±4.66 m.

Investigation of Heterogeneity Sources for Occupational Task Recognition via Transfer Learning

Human activity recognition has been extensively used for the classification of occupational tasks. Existing activity recognition approaches perform well when training and testing data follow an identical distribution. However, in the real world, this condition may be violated due to existing heterogeneities among training and testing data, which results in degradation of classification performance. This study aims to investigate the impact of four heterogeneity sources, cross-sensor, cross-subject, joint cross-sensor and cross-subject, and cross-scenario heterogeneities, on classification performance. To that end, two experiments called separate task scenario and mixed task scenario were conducted to simulate tasks of electrical line workers under various heterogeneity sources. Furthermore, a support vector machine classifier equipped with domain adaptation was used to classify the tasks and benchmarked against a standard support vector machine baseline. Our results demonstrated that the support vector machine equipped with domain adaptation outperformed the baseline for cross-sensor, joint cross-subject and cross-sensor, and cross-subject cases, while the performance of support vector machine equipped with domain adaptation was not better than that of the baseline for cross-scenario case. Therefore, it is of great importance to investigate the impact of heterogeneity sources on classification performance and if needed, leverage domain adaptation methods to improve the performance.