DESIGN AND VALIDATION OF A LOW-COST SONAR PROTOTYPE FOR SYNTHETIC DATASET GENERATION IN UNDERWATER MINE DETECTION

Detection of underwater mines is very important for military, maritime security, and environment safety applications. However, the development of machine learning models is limited heavily because of the lack of quality labeled sonar datasets, especially in military contexts as the data there is highly confidential and expensive. The problem with current synthetic datasets is that they fail to properly replicate how complex operational underwater environments are which leads to major performance gaps when deployed in the real world. This work shows the design and validation of a low-cost sonar prototype, specifically developed for synthetic dataset generation to work on the issue of the scarcity of data in applications of mine detection underwater.
The sonar prototype was built using an Arduino Uno microcontroller, Texas Instruments TUSS4470 ultrasonic analog front end along with a 200khz waterproof transducer in a controlled water tank environment. For echo analysis the system generates 16 cycle bursts and captures approximately 850 samples at 13 μs intervals. The signal processing consists of zero-phase low-pass Butterworth filtering, short-time energy analysis, and adaptive thresholding which is for echo detection. Under varying conditions (like, salinity 0-35 ppt and temperature 10-30°C), the sonar prototype operated successfully and was able to produce high fidelity acoustic datasets. These datasets are suitable for training machine learning models. The sonar prototype provides a proper platform for the generation of acoustic datasets that are realistic under varying environmental conditions and offers a lot of potential for improving the training of machine learning models and generalization in applications of detecting underwater mines.