Instance Segmentation in Very High Resolution Remote Sensing Imagery Based on Hard-to-Segment Instance Learning and Boundary Shape Analysis

Although great success has been achieved in instance segmentation, accurate segmentation of instances remains difficult, especially at object edges. This problem is more prominent for instance segmentation in remote sensing imagery due to the diverse scales, variable illumination, smaller objects, and complex backgrounds. We find that most current instance segmentation networks do not consider the segmentation difficulty of different instances and different regions within the instance. In this paper, we study this problem and propose an ensemble method to segment instances from remote sensing images, considering the enhancement of hard-to-segment instances and instance edges. First, we apply a pixel-level Dice metric that reliably describes the segmentation quality of each instance to achieve online hard instance learning. Instances with low Dice values are studied with emphasis. Second, we generate a penalty map based on the analysis of boundary shapes to not only enhance the edges of objects but also discriminatively strengthen the edges of different shapes. That is, different areas of an object, such as internal areas, flat edges, and sharp edges, are distinguished and discriminatively weighed. Finally, the hard-to-segment instance learning and the shape-penalty map are integrated for precise instance segmentation. To evaluate the effectiveness and generalization ability of the proposed method, we train with the classic instance segmentation network Mask R-CNN and conduct experiments on two different types of remote sensing datasets: the iSAID-Reduce100 and the JKGW_WHU datasets, which have extremely different feature distributions and spatial resolutions. The comprehensive experimental results show that the proposed method improved the segmentation results by 2.78% and 1.77% in mask AP on the iSAID-Reduce100 and JKGW_WHU datasets, respectively. We also test other state-of-the-art (SOTA) methods that focus on inaccurate edges. Experiments demonstrate that our method outperforms these methods.

Global model for uneven illumination enhancement of document images

Exponential growth of low-cost digital imagery is latterly observed. Images acquired under uneven lighting are prone to experience poor visibility, which may severely limit the performance of most computational photography and automatic visual recognition applications. Different from current optimization techniques, we design a novel partial differential equation based model to rectify the variable illumination artifacts. In this study, a large number of document samples capturing uneven illumination and low contrast conditions are tested to compare the effectiveness of the proposed local and nonlocal approaches. The suggested algorithm can be applied in Arabic, Latin and Chinese document images with any type of shade.

Brightness variations of young Sun-like stars from ground-based and space telescopes

Young stellar objects often show photometric variability, which is well examined at optical wavelengths, but more and more infrared data are also available. The wavelength dependence of the variability carries information on the physical cause of the changing brightness. Here, we examine seven T Tauri-type stars known for their large amplitude variability selected from the Campaign 13 field of the Kepler K2 mission. We complemented the K2 light curves by multifilter optical monitoring observations made with the 90 cm Schmidt telescope of Konkoly Observatory, and by 3.6 and 4.5 μm infrared photometry with a 20 hours cadence using the Spitzer Space Telescope. We found that the wavelength dependence of the observed variability is not consistent with changing interstellar extinction. We suggest that the brightness changes are due to variable accretion, causing a variable illumination of the inner disk.