Multi-Class Strategies for Joint Building Footprint and Road Detection in Remote Sensing

Building footprints and road networks are important inputs for a great deal of services. For instance, building maps are useful for urban planning, whereas road maps are essential for disaster response services. Traditionally, building and road maps are manually generated by remote sensing experts or land surveying, occasionally assisted by semi-automatic tools. In the last decade, deep learning-based approaches have demonstrated their capabilities to extract these elements automatically and accurately from remote sensing imagery. The building footprint and road network detection problem can be considered a multi-class semantic segmentation task, that is, a single model performs a pixel-wise classification on multiple classes, optimizing the overall performance. However, depending on the spatial resolution of the imagery used, both classes may coexist within the same pixel, drastically reducing their separability. In this regard, binary decomposition techniques, which have been widely studied in the machine learning literature, are proved useful for addressing multi-class problems. Accordingly, the multi-class problem can be split into multiple binary semantic segmentation sub-problems, specializing different models for each class. Nevertheless, in these cases, an aggregation step is required to obtain the final output labels. Additionally, other novel approaches, such as multi-task learning, may come in handy to further increase the performance of the binary semantic segmentation models. Since there is no certainty as to which strategy should be carried out to accurately tackle a multi-class remote sensing semantic segmentation problem, this paper performs an in-depth study to shed light on the issue. For this purpose, open-access Sentinel-1 and Sentinel-2 imagery (at 10 m) are considered for extracting buildings and roads, making use of the well-known U-Net convolutional neural network. It is worth stressing that building and road classes may coexist within the same pixel when working at such a low spatial resolution, setting a challenging problem scheme. Accordingly, a robust experimental study is developed to assess the benefits of the decomposition strategies and their combination with a multi-task learning scheme. The obtained results demonstrate that decomposing the considered multi-class remote sensing semantic segmentation problem into multiple binary ones using a One-vs.-All binary decomposition technique leads to better results than the standard direct multi-class approach. Additionally, the benefits of using a multi-task learning scheme for pushing the performance of binary segmentation models are also shown.

Binary decompositions of probability densities and random-bit simulation

This paper is devoted to random-bit simulation of probability densities, supported on {[0,1]}. The term “random-bit” means that the source of randomness for simulation is a sequence of symmetrical Bernoulli trials. In contrast to the pioneer paper [D. E. Knuth and A. C. Yao, The complexity of nonuniform random number generation, Algorithms and Complexity, Academic Press, New York 1976, 357–428], the proposed method demands the knowledge of the probability density under simulation, and not the values of the corresponding distribution function. The method is based on the so-called binary decomposition of the density and comes down to simulation of a special discrete distribution to get several principal bits of output, while further bits of output are produced by “flipping a coin”. The complexity of the method is studied and several examples are presented.

Why is the American South Poorer?

ABSTRACT While American poverty research has devoted greater attention to poverty in the Northeast and Midwest, poverty has been persistently higher in the U.S. South than in the other regions. Thus, this study investigates the enduring question of why poverty is higher in the South. Specifically, it demonstrates the role of power resources as an explanation for this regional disparity, yet also considers family demography, economic structure, and racial/ethnic heterogeneity. Using six waves (2000–2016) of U.S. Census Current Population Survey data from the Luxembourg Income Study (N = 1,157,914), this study employs a triangulation of analytic techniques: (1) tests of means and proportion differences, (2) multilevel linear probability models of poverty, and (3) binary decomposition of the South/non-South poverty gap. The comparison of means associated with the power resource hypothesis yields the largest substantive differences between the South and the non-South. In the multilevel models, adjusting for power resources yields the largest declines in the South coefficient. Binary decomposition results indicate power resources are the second most influential factor explaining the South/non-South poverty gap. Overall, power resources are an important source of the South/non-South poverty gap, though economic structure and other factors certainly also play a role. Results also suggest an important interplay between power resources and race. Altogether, these results underscore the importance of macrolevel characteristics of places, including political and economic contexts, in shaping individual poverty and overall patterns of inequality.

Composite Binary Decomposition Networks

Binary neural networks have great resource and computing efficiency, while suffer from long training procedure and non-negligible accuracy drops, when comparing to the fullprecision counterparts. In this paper, we propose the composite binary decomposition networks (CBDNet), which first compose real-valued tensor of each layer with a limited number of binary tensors, and then decompose some conditioned binary tensors into two low-rank binary tensors, so that the number of parameters and operations are greatly reduced comparing to the original ones. Experiments demonstrate the effectiveness of the proposed method, as CBDNet can approximate image classification network ResNet-18 using 5.25 bits, VGG-16 using 5.47 bits, DenseNet-121 using 5.72 bits, object detection networks SSD300 using 4.38 bits, and semantic segmentation networks SegNet using 5.18 bits, all with minor accuracy drops.1

Classification of Thyroid Nodules in Ultrasound Images Using Direction-Independent Features Extracted by Two-Threshold Binary Decomposition

In recent years, several computer-aided diagnosis systems emerged for the diagnosis of thyroid gland disorders using ultrasound imaging. These systems based on machine learning algorithms may offer a second opinion to radiologists by evaluating a malignancy risk of thyroid tissue, thus increasing the overall diagnostic accuracy of ultrasound imaging. Although current computer-aided diagnosis systems exhibit promising results, their use in clinical practice is limited. One of the main limitations is that the majority of them use direction-dependent features. Our intention has been to design a computer-aided diagnosis system, which will use only direction-independent features, that is, it will not be dependent on the orientation and the inclination angle of the ultrasound probe when acquiring the image. We have, therefore, applied histogram analysis and segmentation-based fractal texture analysis algorithm, which calculates direction-independent features only. In our study, 40 thyroid nodules (20 malignant and 20 benign) were used to extract several features, such as histogram parameters, fractal dimension, and mean brightness value in different grayscale bands (obtained by 2-threshold binary decomposition). The features were then used in support vector machine and random forests classifiers to differentiate nodules into malignant and benign classes. Using leave-one-out cross-validation method, the overall accuracy was 92.42% for random forests and 94.64% for support vector machine. Results show that both methods are useful in practice; however, support vector machine provides better results for this application. Proposed computer-aided diagnosis system can provide support to radiologists in their current diagnosis of thyroid nodules, whereby it can optimize the overall accuracy of ultrasound imaging.

Existence of Prograde Double-Double Orbits in the Equal-Mass Four-Body Problem

By introducing simple topological constraints and applying a binary decomposition method, we show the existence of a set of prograde double-double orbits for any rotation angle {\theta\in(0,\pi/7]} in the equal-mass four-body problem. A new geometric argument is introduced to show that for any {\theta\in(0,\pi/2)}, the action of the minimizer corresponding to the prograde double-double orbit is strictly greater than the action of the minimizer corresponding to the retrograde double-double orbit. This geometric argument can also be applied to study orbits in the planar three-body problem, such as the retrograde orbits, the prograde orbits, the Schubart orbit and the Hénon orbit.