@phdthesis{oai:kyutech.repo.nii.ac.jp:00004074,
 author = {Tan, Yasuhiro and 丹, 康弘},
 month = {2014-07-30},
 note = {Chapter 1 Introduction||Chapter 2 Underwater acoustics||Chapter 3 Detection of underwater objects based on machine learning||Chapter 4  Automatic classification of seabed sediments using HLAC||Chapter 5 Conclusion, The ocean accounts for approximately 70% of the area on the earth, and the water as well as coastal areas sustain many species including humans. Ocean resources are used for fish farming, land reclamation, and a variety of other purposes. Seabed resources such as oil, natural gas methane hydrates, and manganese nodules are still largely unexploited on the bottom of the sea. Maps are critical to development activities such as construction, mining, offshore drilling, marine traffic control, security, environmental protection, and tourism. Accordingly, more topographic and others types of mapping information are needed for marine and submarine investigations. Both waterborne and airborne survey techniques show promise for collecting data on marine and submarine environments, and these techniques can be classified into four main categories. First, remote sensing by satellites or aircraft is a widely used technique that can yield important data such as information on sea levels and coastal sediment transport. Second, investigations may collect direct information by remotely operated vehicles (ROVs), autonomous underwater vehicles (AUVs), and divers. While the quality of data obtained from these techniques is high, the data obtained are often limited to relatively shallow and small geographic areas. Third, sediment profile imagery can be used to collect photographs that contain detailed information about the seabed. Lastly, acoustic investigations that use sonar are popular in marine mapping studies, especially in coastal areas. In particular, acoustic investigations that employ ultrasound technology can yield rich information about variations in bathymetry.
Unlike air, water has physical properties that make it difficult for light or electromagnetic waves to pass through. However, sound waves propagate readily in water. Therefore, sound waves are used in a wide range of technical applications to detect underwater structures that are difficult to observe with light-based techniques. In the dark depths of the ocean, the use of acoustic technology is essential. The development of marine acoustic technology is expanding in modern times. In addition to the basic physics related to acoustic waves, much research has been dedicated to other basic and applied fields such as electronics, physical oceanography, signal processing, and biology. The realization of new sonar systems that utilize advanced detection algorithms can be expected to contribute to major breakthroughs in oceanographic research that require deployment to novel marine environments and other areas of natural resource interest.
In this study, the author focuses on side-scan sonar, which is one of the imaging technologies that employs sound to determine the seabed state, to conduct research on imaging algorithms for discrimination. The proposed method for discrimination was coupled to a high-speed detection method for installed reefs on the seabed. This method is also capable of detecting unknown objects with Haar-like features during object recognition of rectangular regions of a certain size via machine learning by AdaBoost and fast elimination of non-object regions on the cascade structure. Side-scan and forward looking sonars are some of the most widely used imaging systems for obtaining large-scale images of the seafloor, and their application continues to expand rapidly with their increasing deployment on AUVs. However, it can be difficult to extract quantitative information from the images generated from these processes, in particular, for the detection and extraction of information on the objects within these images. Hence, this study analyzes features that are common to most undersea objects projected in side-scan sonar images to improve information processing. By using a technique based on the k-means method to determine the Haar-like features, the number of patterns of Haar-like features was minimized and the proposed method was capable of detecting undersea objects faster than current methodology. 
This study demonstrates the effectiveness of this method by applying it to the detection of real objects imaged on the seabed (i.e., sandy ground and muddy ground). 
Attempts are made as well to automate the proposed method for discriminating objects lying on the seafloor from surficial sediments. During undersea exploration, a thorough understanding of the state of the seafloor surrounding objects of interest is important. Therefore, a method is proposed in this study to automatically determine seabed sediment characteristics. In traditional methods, a variety of techniques have been used to collect information about seabed sediments including depth measurements, bathymetry evaluations, and seabed image analyses using the co-occurrence direction of the gray values of the image. Unfortunately, such data cannot be estimated from the object image itself and it can take a long time to obtain the required information. Therefore, these techniques are not currently suitable for real-time identification of objects on the seafloor. For practical purposes, automatic techniques that are developed should follow a simple procedure that results in highly precise and accurate classifications. The technique proposed here uses the subspace method, which is a method that has been used for supervised pattern recognition and analyses of higher-order local autocorrelation features. The most important feature of this method is that it uses only acoustic images obtained from the side-scan sonar. This feature opens up the possibility of installing this technology in unmanned small digital devices. In this study, the classification accuracy of the proposed automation method is compared to the accuracy of traditional methods in order to show the usefulness of the technology. In addition, the proposed method is applied to real-world images of the seabed to evaluate its effectiveness in marine surveys.
The thesis is organized as follows. In Chapter 1, the purpose of this study is presented and previous studies relevant to this research are reviewed. In Chapter 2, an overview of underwater sound is given and key principles of sound wave technology are explained. In Chapter 3, a new method for detecting and discriminating objects on the seafloor is proposed. In Chapter 4, the possibility of automating the discrimination method is explored. Finally, Chapter 5 summarizes the findings of this study and proposes new avenues for future research., 九州工業大学博士学位論文 学位記番号:工博甲第364号　学位授与年月日:平成26年3月25日, 平成25年度},
 school = {九州工業大学},
 title = {A Study on the Recognition of Seabed Environments Employing Sonar Images},
 year = {}
}