A Research On Underwater Acoustic Image Feature Extraction

With the development of the acoustic imaging technology, sonar image is being widely used in ocean exploitation. The target identification based on sonar image has become an important subject in the image processing domain. As an important part of the target identification, feature extraction of sonar image is developing faster and faster. This thesis deeply studies on feature extraction based on texture and shape feathers of acoustic images. The main contents are showed as follows:1. According to the sonar images which have simple background and clear shapes, the thesis extracts the classical shape features as perimeter, area, form parameter, major and minor axes and eccentricity. This feature vectors can be used to classify and recognize the underwater target.2. According to characters of the sonar images, this thesis extracts the shape features as regional moment invariants, edge moment invariants and form marking function moment invariants. As these features have the rotation invariance, scale invariance and transfer invariance, so they can be used to recognize the underwater targets. Experiment is done to analysis and compare the advantages and disadvantages of the three methods in recognition speed and capability.3. A neighborhood gray level co-occurrence matrix is confirmed for performing texture analysis. Feature vector of six dimensions are used to classify three different types of seafloor sediment images (sand, rock, ice).4. To deal with the problem of seabed reverberation of sonar images in target detection, a method is studied to estimate the target in images by the magnitude of higher-order spectrum, which is called bi-spectrum. This method can restrain the background noise in images and enhanced the target information, so it can detect target in different side-scan sonar images effectively and universally.