Seismic inversion with deep learning

This article investigates bypassing the inversion steps involved in a standard litho-type classification pipeline and performing the litho-type classification directly from imaged seismic data. We consider a set of deep learning methods that map the seismic data directly into litho-type classes, trained on two variants of synthetic seismic data: (i) one in which we image the seismic data using a local Radon transform to obtain angle gathers, (ii) and another in which we start from the subsurface-offset gathers, based on correlations over the seismic data. Our results indicate that this single-step approach provides a faster alternative to the established pipeline while being convincingly accurate. We observe that adding the background model as input to the deep network optimization is essential in correctly categorizing litho-types. Also, starting from the angle gathers obtained by imaging in the Radon domain is more informative than using the subsurface offset gathers as input.

Saliency Detection with Moving Camera via Background Model Completion

Detecting saliency in videos is a fundamental step in many computer vision systems. Saliency is the significant target(s) in the video. The object of interest is further analyzed for high-level applications. The segregation of saliency and the background can be made if they exhibit different visual cues. Therefore, saliency detection is often formulated as background subtraction. However, saliency detection is challenging. For instance, dynamic background can result in false positive errors. In another scenario, camouflage will result in false negative errors. With moving cameras, the captured scenes are even more complicated to handle. We propose a new framework, called saliency detection via background model completion (SD-BMC), that comprises a background modeler and a deep learning background/foreground segmentation network. The background modeler generates an initial clean background image from a short image sequence. Based on the idea of video completion, a good background frame can be synthesized with the co-existence of changing background and moving objects. We adopt the background/foreground segmenter, which was pre-trained with a specific video dataset. It can also detect saliency in unseen videos. The background modeler can adjust the background image dynamically when the background/foreground segmenter output deteriorates during processing a long video. To the best of our knowledge, our framework is the first one to adopt video completion for background modeling and saliency detection in videos captured by moving cameras. The F-measure results, obtained from the pan-tilt-zoom (PTZ) videos, show that our proposed framework outperforms some deep learning-based background subtraction models by 11% or more. With more challenging videos, our framework also outperforms many high-ranking background subtraction methods by more than 3%.

Fitting electron spectrum from AMS-02 by pulsar electrons

In this work, we aim to explain the latest data of cosmic-ray electrons from AMS-02 by an electron background model and pulsar electrons. We consider an electron background model which includes primary and secondary electrons. We assume that pulsars are major sources of the electron excess. Since electrons easily lose their energy through the interstellar radiation field and the magnetic field via inverse Compton scattering and synchrotron radiation, respectively, they propagate in a short length. We adopt nearby pulsar data in the distance of 1 kpc from the Australia Telescope National Facility (ATNF) pulsar catalogue. By using a Green’s function of an electron propagation model, we then fit pulsar parameters (i.e. the spectral index, the fraction of the total spin-down energy and the cutoff energy) for several cases of a single pulsar. With a combination of the electron background model, several cases of pulsar spectrum are able to explain the electron excess.

Detection for Dangerous Goods Vehicles in Expressway Service Station Based on Surveillance Videos

To improve the safety capabilities of expressway service stations, this study proposes a method for detecting dangerous goods vehicles based on surveillance videos. The information collection devices used in this method are the surveillance cameras that already exist in service stations, which allows for the automatic detection and position recognition of dangerous goods vehicles without changing the installation of the monitoring equipment. The process of this method is as follows. First, we draw an aerial view image of the service station to use as the background model. Then, we use inverse perspective mapping to process each surveillance video and stitch these videos with the background model to build an aerial view surveillance model of the service station. Next, we use a convolutional neural network to detect dangerous goods vehicles from the original images. Finally, we mark the detection result in the aerial view surveillance model and then use that model to monitor the service station in real time. Experiments show that our aerial view surveillance model can achieve the real-time detection of dangerous goods vehicles in the main areas of the service station, thereby effectively reducing the workload of the monitoring personnel.

Joint inversion of the reflectivity and the velocity model

During linearized waveform inversion, the presence of small inaccuracies in the background subsurface model can lead to unfocused seismic events in the final image. The effect on the amplitude can mislead the interpretation. We present a joint inversion scheme in the model domain of the reflectivity and the background velocity model. The idea is to unify the inversion of the background and the reflectivity model into a single framework instead of treating them as decoupled problems. We show that with this method, we can obtain a better estimate of the reflectivity than that obtained with conventional linearized waveform inversion. Conversely, the background model is improved by the joint inversion with the reflectivity in comparison with wave-equation migration velocity analysis. We perform tests on 2D synthetics and 3D field data that demonstrate both benefits.

NWP verification against own analysis by using a Data Assimilation confidence mask

<p>Numerical Model Prediction (NWP) verification against station measurements from a surface network is affected by sub-tile representativeness issues. Moreover, the station network is often not representative of the whole verification domain (e.g. usually coastal stations are predominant) and large unpopulated regions (such as oceans, Polar regions, deserts) are under-sampled. Verification against gridded analyses mitigate these issues, since they partially address the sub-tile representativeness, and sample homogeneously the verification domain. Moreover, gridded analyses merge station network measurements to radar and satellite retrieval estimates, in a physical coherent fashion, over the same NWP grid. Verification against own analysis, despite quite convenient, is however hampered by its dependence on the NWP background model, which renders the verification “incestuous”, further than being affected by the uncertainties introduced by retrieval algorithms and Data Assimilation (DA) procedures.</p><p>In this study we investigate the use of a gridded NWP own analysis for verification, by applying a mask to reduce the background model contribution. The mask weights the verification scores to account for the amounts of observations assimilated and their associated uncertainty, as estimated from DA. We illustrate the approach by using the Canadian Precipitation Analysis (CaPA), which assimilates station measurements, radar and satellite-based (IMERG) observations. The CaPA confidence (weighting) mask is dynamic and changes depending on the daily available (assimilated) observations, and on their corresponding DA error statistics; it is defined as</p><p>                                             mask = 1 - var(A-O)/var(B-O)</p><p>where A=analysis, B=Background, O=observations. Where the analysis is identical to the background model, the weighting mask is zero.</p><p>We evaluate the Canadian Regional Deterministic Prediction System (RDPS), which is the NWP system used as background model for CaPA. As expected, the verification results obtained by using the weighting mask lay between the verification results obtained verifying against the analysis over the full domain, and the results obtained verifying against station measurements. The effects of sub-tile representativeness are quantified by comparing verification results against station measurements to verification results against CaPA for the grid-points co-located with the stations. Finally, the comparison of the verification results against CaPA over the full domain versus the verification results against CaPA for the grid-points co-located with stations, estimates to which extent the station network is representative of the full domain.</p><p>The approach aims to propose a simple -yet effective- better practice for verification against own analysis.</p>

Binary background model with geometric mean for author-independent authorship verification

Authorship verification (AV) is one of the main problems of authorship analysis and digital text forensics. The classical AV problem is to decide whether or not a particular author wrote the document in question. However, if there is one and relatively short document as the author’s known document, the verification problem becomes more difficult than the classical AV and needs a generalised solution. Regarding to decide AV of the given two unlabeled documents (2D-AV), we proposed a system that provides an author-independent solution with the help of a Binary Background Model (BBM). The BBM is a supervised model that provides an informative background to distinguish document pairs written by the same or different authors. To evaluate the document pairs in one representation, we also proposed a new, simple and efficient document combination method based on the geometric mean of the stylometric features. We tested the performance of the proposed system for both author-dependent and author-independent AV cases. In addition, we introduced a new, well-defined, manually labelled Turkish blog corpus to be used in subsequent studies about authorship analysis. Using a publicly available English blog corpus for generating the BBM, the proposed system demonstrated an accuracy of over 90% from both trained and unseen authors’ test sets. Furthermore, the proposed combination method and the system using the BBM with the English blog corpus were also evaluated with other genres, which were used in the international PAN AV competitions, and achieved promising results.