Artificial Intelligence Based Multi-modal Sensing for Flash Flood Investigation

Abstract

Flash floods are very abrupt and sudden and can devastate large areas within a fraction of seconds. Tsunami is also a grass root cause of infrastructure loss and casualties. Tsunami is caused by the release of energy inside the ocean. It has been observed that Tornados and Tsunami frequently occurred that caused more than 120000 casualties between 1992 to 2005. Several approaches and cases have been studied and applied to overcome this topmost issue. Strong and reliable flood risk management system must have the capabilities to forecast and estimate the change in the process: the hydro atmospheric and climatic development that causes flood waves and land and metropolitan exposure that leads to casualties. Techniques utilized for the early investigation of flash floods may be classified into the following categories a. Sensors and instrumentation-based b. Radar-based c. Satellite images based. In this paper, researchers presented a reliable and vigor cost-effective solution for the prediction of flash floods accurately and precisely by using direct measurements with the combination of scaled conjugate gradient back propagation. Researchers have adopted direct measurement method by utilizing a couple of sensors named as PIR, ultrasonic sensor, water or humidity sensor, temperature and pressure sensors. Gas sensor has also been used to detect the increased carbon dioxide levels in the environment as the soil is saturated more and the probability of flash flood occurrence becomes sure in this situation. An appropriate mixture of estimation sensors can extensively boost the benefit of information in contrast with that from a single sensor. 48 hours of data have been recorded and processed for the false alarm identification by using Scaled conjugate gradient back propagation. Results proved that our proposed system worked better or equivalent to the other available techniques with minimized error rate. Results showed that it performed better than the existing approaches. Multi-layer perceptron (MLP) was applied to the same data set for the error and false alarm classification. Results were compared with the multi-layer perceptron and it can be easily observed that they are close enough, therefore, scaled conjugate gradient back propagation performed very well.