Cellular neural networks for real-time monitoring of volcanic activity

The paper introduces a new methodology for real-time monitoring of active v olcanoes, which is based on efficient video processing operations implement ed by means of cellular neural network (CNN) architectures. CNNs are massiv e parallel analog circuits with only local interconnections between the com puting elements, that are programmed in an analog way to perform almost all image processing operations. The performance of CNN-based operations is re ported by simulation of some dynamic image processing tasks in active volca no monitoring. The purpose of the proposed computer-based system for volcan ic image processing is twofold: on-line signalling of volcanic events of in terest such as lava fountains, Strombolian explosions, ash and gas emission s, etc., and real-time extraction of quantitative information which charact erises the events, i.e. geometric parameters, energy involved, type of even t and so on. The performance of the present version of the system is limite d, in terms of processing speed, by the simulator instead of the on-chip an alog CNN, which is still under development by STMicroelectronics, Hence the system can operate well only when volcanic activity is not paroxysmal. The system has been tested on images taken both on Etna and Stromboli, volcano es located in southern Italy, but it can easily be adapted in order to work in other volcanic areas. The technique implemented for the image-processing operations, called 'CNN- ADI', was conceived for moving image processing and combines the cumulative differences model with the computational speed and versatility of CNNs, im plementing a pseudo-ADI (accumulative difference image) algorithm. The adva ntage of using a CNN-based version of the ADI filter lies in the possibilit y of real-time filtering, directly on-chip of short sequences of images to distinguish between the dynamic and static elements the frames contain. The main advantages of the present work are that not only are human operato rs relieved of the task of visual monitoring but it is also possible to ext ract on-line physical parameters of volcanic events, including event classi fication. (C) 1999 Elsevier Science Ltd. All rights reserved.