Sound Recognition Technology For Helicopter Using Vector Quantization

In1939, the first helicopter VS-300was born in USA, along with the development of the world and the modern war, the helicopter as an air combat platform, is playing an important role. Because of the structural features and technical characteristics of helicopter, it can flight in the low altitude. It is the important transport tool and fire strike mean in the battlefield, has become an important hub connecting land, sea and air battle. But on the other hand, no matter what kind of attitude helicopter in flight, it will produce a sound signal that is not easy to eliminate. The technology of sound recognition for helicopter is to analyze the sound, obtain important information, and achieve the recognition, location and tracking for the helicopter. Therefore, research of sound recognition technology for helicopter has a very important practical significance.In this thesis, we describe the causes and classification of the helicopter sound, and compare it with the speech signal. Research shows that, the helicopter sound and speech signal is the same in recognition, so the analysis method in speech recognition applied to the sound recognition of helicopter is feasible and effective. In the field of speech recognition, we introduces principles and steps of recognition, and introduces the key technologies, including preprocessing, feature extraction and recognition.Currently, commonly algorithm used in speech recognition are Dynamic Time Warping (DTW), Vector Quantization (VQ), Hidden Markov Model (HMM) and Artificial Neural Network (ANN), etc. Vector Quantization is one of the matching technology which has a high performance.This thesis focuses on the recognition technology based on vector quantization. The classic LBG algorithm is easy to fall into a local extreme value, and the initial codebook has great impact on the optimal codebook in the process of designing a codebook, Quantum Genetic Algorithm (QGA) is a hot interdisciplinary research in genetic algorithm, which is a new searching algorithm that combined quantum parallel computing ability with genetic algorithm optimization ability. Compared with GA, QGA has the merits of large searching range, flexible adaptability and high efficiency. In this thesis we apply QGA to LBG to improve its performance and obtain a new algorithm, which is called QGA-LBG. We apply QGA-LBG to helicopter sound recognition, and perform simulation in MATLAB, to compared QGA-LBG with the classical LBG and GA-LBG. Experimental results show that QGA-LBG can improve the recognition rate of helicopter sound. Moreover, the recognition ratio of QGA-LBG will increase significantly than that of LBG and GA-LBG, if we increase the length of training sound.