Application of Support Vector Machines for Seismogram Analysis and Differentiation
Authors:
Rohit Kumar Shrivastava
Abstract:
Support Vector Machines (SVM) is a computational technique which has been used in various fields of sciences as a classifier with k-class classification capability, k being 2,3,4, etc. Seismograms of volcanic tremors often contain noises which can prove harmful for correct interpretation. The PCAB station (located in the northern region of Panarea island, Italy) has been recording seismic signals…
▽ More
Support Vector Machines (SVM) is a computational technique which has been used in various fields of sciences as a classifier with k-class classification capability, k being 2,3,4, etc. Seismograms of volcanic tremors often contain noises which can prove harmful for correct interpretation. The PCAB station (located in the northern region of Panarea island, Italy) has been recording seismic signals from a pump installed nearby, corrupting the useful signals from Strombolli volcano. SVM with k=2 classification technique after optimization through grid search has been instrumental in identification and classification of the seismic signals coming from pump, reaching a score of 99.7149% of patterns which match the actual membership of class (determined through cross-validation). The predicted labels of SVM has been used to estimate the pump's duration of activity leading to the declaration of corresponding seismograms redundant (not fit for processing and interpretation). However, when the same trained SVM was used to determine whether the seismogram used by Pino et al., 2011 recorded at the same PCAB station on 4th April, 2003 contained pump's signals or not, SVM showed 100% absence of pump's signals thereby authenticating the research work done in the latter.
△ Less
Submitted 5 March, 2020;
originally announced March 2020.
Unsupervised Automated Event Detection using an Iterative Clustering based Segmentation Approach
Authors:
Deepak K. Gupta,
Rohit K. Shrivastava,
Suhas Phadke,
Jeroen Goudswaard
Abstract:
A class of vision problems, less commonly studied, consists of detecting objects in imagery obtained from physics-based experiments. These objects can span in 4D (x, y, z, t) and are visible as disturbances (caused due to physical phenomena) in the image with background distribution being approximately uniform. Such objects, occasionally referred to as `events', can be considered as high energy bl…
▽ More
A class of vision problems, less commonly studied, consists of detecting objects in imagery obtained from physics-based experiments. These objects can span in 4D (x, y, z, t) and are visible as disturbances (caused due to physical phenomena) in the image with background distribution being approximately uniform. Such objects, occasionally referred to as `events', can be considered as high energy blobs in the image. Unlike the images analyzed in conventional vision problems, very limited features are associated with such events, and their shape, size and count can vary significantly. This poses a challenge on the use of pre-trained models obtained from supervised approaches.
In this paper, we propose an unsupervised approach involving iterative clustering based segmentation (ICS) which can detect target objects (events) in real-time. In this approach, a test image is analyzed over several cycles, and one event is identified per cycle. Each cycle consists of the following steps: (1) image segmentation using a modified k-means clustering method, (2) elimination of empty (with no events) segments based on statistical analysis of each segment, (3) merging segments that overlap (correspond to same event), and (4) selecting the strongest event. These four steps are repeated until all the events have been identified. The ICS approach consists of a few hyper-parameters that have been chosen based on statistical study performed over a set of test images. The applicability of ICS method is demonstrated on several 2D and 3D test examples.
△ Less
Submitted 22 January, 2019;
originally announced January 2019.
