Research On Image Building And Features Extraction Method Of AUV Forward-looking Sonar

With the development of ocean exploration, the demand of underwater vehicle with the capability of autonomous navigation is growing. In the complex underwater environment, the sonar sensor is perceived as important equipment of autonomous underwater vehicle (AUV). Underwater environmental exploration, features extraction and targets recognition are perceived as the core issue in the field of AUV. Contrasts to traditional sonar system, there must have the functions of environmental perception, information acquisition, sonar imaging, features extraction and recognition in the AUV acoustic visual system. Therefore, it appears to be important to research on this.This paper is based on the National High-tech Research Development Plans (863 plans): Research on Key technology of AUV integrated navigation system based on sonar and underwater vision under deep-sea complex environment. Its main work is the further study of the AUV underwater acoustic detection and navigation system, and to find the related techniques suitable for AUV sonar image processing, features extraction and recognition.In sonar imaging, the paper first introduces the use of Super Seaking DST double-frequency digital mechanical scanning forward-looking sonar from the Tritech company. It is used as the main detecting and perception tool of AUV to receive the echo signals from underwater environment. In order to realize underwater environment visualization, the signals are effectively selected and processed to build the original image of underwater environment. Among them, the situation of collecting the echo data during the process of AUV motion is discussed. And the effective compensation method is proposed.In sonar image processing, this paper first introduces and analyzes the structure models, the basic principles and applications in image processing of the pulse coupled neural network (PCNN) and intersecting cortical model (ICM). Among them, the necessary contrast and analysis is presented. Then, it proposes the image preprocessing based on ICM neuron models such as contrast transform and image segmentation, which increases the continuity, real-time and effectiveness of the preprocessing and features extraction phases.In features extraction, this paper introduces the classical geometry invariant features, and moment features with translation, rotation and scale invariance which are served as the feature vectors to describe targets. This paper studies a kind of calculation framework about improved image geometric feature extraction. The framework combines vertex chain code with the discrete green theorem and optimizes the calculation of feature vectors. It can quickly calculate target feature vectors, and can be effectively applied in subsequent object recognition and tracking. Then, in view of large closed plaques whose returns are relatively gentle in underwater environment, this paper analyzes and contrasts the least square basic theory and the application of elliptic fitting, likewise, calculates torque characteristic value and extracts the elliptic features of regional plaques using this calculation framework. Finally, combining with the image features extraction, recognition technology and sonar image characteristics summarized in the previous chapters, this paper proposes and realizes sonar image processing and features extraction method suitable for AUV.