Incremental Data Stream Mining and Conflict Analysis for the Recognition of Sonar Signals Underwater

Abstract

Sound navigation and range, or sonar, is a sound propagation technique that is used for underwater navigation, communication, and/or detection of sub marine objects. According to [1], the relevant methodologies have recently been examined. The detection and categorization of sonar sounds, in particular, was cited as one of the most difficult problems in the area. For the detection of items of interest beneath the sea, selecting the proper categorization model for sonar signals identification is critical. Ocean sampling networks, environmental monitoring, offshore exploration, catastrophe prevention, aided navigation, and mine reconnaissance may all benefit from underwater sensor networks. Sensor networks in the sea are simple to set up, don't need any connections, and don't get in the way of shipping. However, long-distance underwater sonar transmissions are susceptible to interference and noise. As a result, sonar signal identification uses data mining methods such as classification to identify the target object's surface from which sonar waves are reflected [3–5]. It is possible to gain significant accuracy in conventional data mining by utilising the whole dataset to build a classification model. Although the induction is normally done and repeated in batches, this means that the model's accuracy is likely to degrade between updates [6]. When new data is added to a dataset, the update time may increase as the total amount of data grows. Sonar signals are relentless and continuously detected, much like any other data stream. Batch mode classification techniques, despite their accuracy, may not be ideal for streaming applications like sonar sensing. It is critical for real-time sensing and reconnaissance to make data processing times as fast as possible since sonar signal data streams might potentially add up to infinity. Using quick conflict analysis from the stream-based training dataset, we provide an alternative data stream mining approach for progressively purging noisy data. With the use of conflict analysis (iDSM-CA), it is known as an incremental data stream mining technique (in acronym). An benefit of this strategy is that it can progressively develop a classification model from stream data. Simulation tests are used to demonstrate the effectiveness of the suggested technique, particularly when it comes to reducing noise from the sonar data while it is streaming. What follows is an outline of the remainder of the paper. Section 2 provides an overview of some of the most often used computational methods for removing noise from training datasets. The "conflict analysis" technique for deleting incorrectly categorised occurrences is described in Section 3 of our novel data stream mining approach. A set of sonar recognition tests are presented in Section 4 to verify the stream mining technique. The paper comes to a close in Section 5.