Sea Technology

DEC 2013

The industry's recognized authority for design, engineering and application of equipment and services in the global ocean community

Issue link: http://sea-technology.epubxp.com/i/226449

Contents of this Issue

Navigation

Page 46 of 83

Hydrographic Equipment & Software Hemisphere GPS Image recognition of Mnemiopsis leidyi captured by Prototype 2 in the Ligurian Sea (a, b, d, e), and image recognition of Sanderia malayensis captured by Prototype 1 at the Aquarium of Genoa (c, f). in front of Cannes and Toulon, France, in collaboration with the JellyWatch monitoring French regional program in spring and summer 2012. Image Analysis Component Many approaches for automatic shape classifcation have been investigated in the literature of computer vision and computer graphics, and some approaches have been tested for marine applications, for instance zooplankton recognition. In general zooplankton recognition has been obtained by frst identifying the region of interest (ROI) within an image by using binarization and segmentation routines. The obtained ROI is then analyzed with support vector machines, decision trees and neural networks. However, no approach has been designed for a stand-alone instrument used in long-lasting deployments on autonomous platforms. In order to meet these requirements, we investigated the method proposed in "Rapid Object Detection Using a Boosted Cascade of Simple Features," by P. Viola and M. Jones. This machine-learning approach produces a simple and computationally light image classifer, which is capable of identifying jellies within the input images. The large number of species, the nonrigid shape and the almost trans- parent body of the gelatinous plankton specimens make image recognition quite challenging. Moreover, the particular acquisition context, light conditions and water turbidity may strongly affect the image quality and thus the recognition performance. Two main experiments have been performed to test the pattern recognition module based on the two sets of images acquired in the Ligurian Sea and in the Aquarium of Genoa. During the experiment in the Ligurian Sea, more than 4,000 images were acquired and 26 specimens captured, which were distributed among 24 images. All of these images are characterized by a uniform background, mainly consisting of seawater, and are representative of what we can expect in open-ocean applications. A reliable statistic cannot be computed because of the small number of acquired specimens, but the results are overall very promising. The correct detection of specimens varies from 69 to 80 percent by varying the algorithm parameters. However, improving the detection results increases the number of false positives (e.g., foating objects mistakenly identifed as jellies). Concerning the data set acquired at the Aquarium of Genoa, a much greater number of specimens were available. Nevertheless, the size and www.sea-technology.com And the following are on GSA! Teledyne ODOM Echo Sounders & Multibeam Systems TELEDYNE TSS Teledyne TSS Motion & Navigation Systems Chesapeake Technology SonarWiz EdgeTech Marine Side Scan Sonars, Subbottom Profilers & Integrated Systems December 2013 / st 47

Articles in this issue

Links on this page

Archives of this issue

view archives of Sea Technology - DEC 2013
loading...
Sea Technology
Welcome!
If you're not a subscriber, please click here.