Research On Sonar Image Feature Extraction

This paper mainly discusses the feature extraction of sonar image. The study was conducted from perspectives of the preprocessing of image, feature extraction of target and classified identification of target.The first part is the preprocessing of sonar image. It has realized the denoising and segmentation of sonar image. Study concerning denoising is the basic sonar image division,including median filtering method, the neighborhood averaging method, wavelet denoising method and morphological smoothing method. Images, respectively for ship propeller frogman and fish image noise removal, image based on different noise conditions, all kinds of denoising method of performance analysis to select suitable for sonar image denoising method. In this step segmentation, this paper studies the iteration method, maximum entropy method and the between-cluster variance method.The second part is part feature extraction. Study is conducted from two aspects, namely,the shape and area feature. Extraction of shape feature comprises the shape of the basic characteristics and the characteristics of the shape based on context. A method of false method removing and region merging was designed and a method of a long axis target endpoint for reference point shape context method. With regards to regional feature extraction, this paper designed a target area feature extraction method using a PCNN neuron neighborhood activation rate and space discrete degree.