Object characterisation using wideband sonar pulses
MetadataShow full item record
Characterisation of objects in an underwater environment is challenging. Success in the task can be beneﬁcial in a variety of scenarios, which include oil and gas pipe maintenance, archaeology, and assistance to general underwater object identiﬁcation. This work focuses on object characterisation, providing a solution for material identiﬁcation. To do this, one must sense the underwater environment for which there are several different ways. Some of the most popular rely on sonar images. These provide limited information about the objects,mostly the shape, size and distance to the object. The study of acoustic wave scattering over a wide frequency range provides more information about the targets characteristics. This work builds on the principles of sound scattering. An acoustic echo reﬂected from an object has a different pulse shape and frequency composition than its initial pulse. These changes in the pulse are due to the interaction of the sound wave with an object during the reﬂection process and the pulses interaction with the transmission medium. Study of the reﬂected pulse can provide information about physical properties such as size, material and shell thickness. The objects used in this work are limited to spherical shells made of a variety of materials, and ﬁlled with different liquids or air. The task of material identiﬁcation is approached in two different ways. The ﬁrst one is a machine learning based approach. The classiﬁcation is not based on the object’s shape, but on its physical properties including the composition material. Two approaches will be presented: one, where the spherical shell is described by the echo’s representation in time frequency domain and one, where it is described by the form function. The objects are classiﬁed using a number of machine learning techniques including support vector machine, gradient boosting and neural networks. The machine learning approaches give good results for a number of tasks, but are not sufﬁcient to distinguish between materials with similar properties, like water and salt water. An alternative solution is presented in this thesis, which identiﬁes the ﬁller and the shell materials separately. This material identiﬁcation approach is based on the timing of the sound scattering components. The echo reﬂected from an object is formed by a number of processes. The information about these processes can be extracted from the echoes and used to identify the material. This approach does not require any training data and shows good results, which are demonstrated on both the simulated and experimental data. This work focuses on object characterisation, providing a solution for material identiﬁcation using underwater acoustics and signal processing techniques.