Real-Time Recognition of Underwater Sonar Signals Using Incremental Stream Mining and Conflict Detection

Abstract

Recognizing sonar sounds is a crucial step in finding large items submerged in the ocean. Rather than relying on visual cues, the military often uses sonar signals to guide them through the depths of the ocean and/or detect the presence of enemy submarines. In particular, data mining's classification technique has been used to sonar signal detection to determine the nature of reflected surfaces. By training a classification model using the whole dataset in batches, classification algorithms in the conventional data mining technique provide reasonable accuracy. It's well knowledge that data streams are continuously acquired from sonar waves.
The prior classification techniques may not be applicable to incremental classifier learning, notwithstanding their efficacy in conventional batch training. To meet the need for fast speed, data preparation time must be minimized despite the infinitely large data streams that might result from sonar signals. To avoid having to learn everything about a dataset all at once, this study introduces a new approach to data mining that is well-suited to the gradual elimination of noisy data through quick conflict analysis of the data stream. Through rigorous simulation studies, we find strong evidence for the methodology's success.