Intrusion Detection Method Research Based On The Characteristics Of The Ultrasonic Echo Envelope

With the development of science and technology, Automatic Teller Machine (ATM) has been widely used in large streets and small lanes, it brings a big convenience to our daily life. In the mean time, it brings a lot of security risk for our operators because of the ATM machine installed outside, it has been solved with the design of safety protection cabin in the ATM and makes a largely protection for our operators'safety. The principle of safety cabin uses infrared detector to test anybody there. In the real application we find that the infrared detector is easy to be affected by external factor (like sun light, temperature, the location of operator stand in cabin etc) and happen to false alarm and miss alarm, so this paper use the character of ultrasonic which is hard to be interrupted by external environment to improve the performance of detector. However, the existed ultrasonic detector judge the invasion by detecting the Doppler frequency of moving objects basically, it is not suitable for the application field described in this paper. So, this paper proposed one method based on echo envelop character to identify the invasion and it can identify the action of the intruder who is moving or stationary.The background of this paper is based on the detection of the safety cabin when it is intruding, it studies and verifies from extracting and identifying the signal character of ultrasonic echo.At first, choose the character. This paper choose the envelope signal as the original character of echo signal. With many observation and analysis to the echo signal and extract the two characteristic quantity of envelope——amplitude character and wavelet transformation——as the final character vector of echo signal, and use the class separability criterion to verify the validity of this character to class.Second, regarding to the identity method, this paper use the k-nearest neighbor (k-NN) and RBF neural network (RBFNN) which are usually used in Pattern Recognition. With the further study on neural network, it proposed an improved RBFNN training method based on the drawback of normal RBFNN training, and realize the improved RBF training method on the Matlab environment. It indicates many merit of the improved training algorithm such as high speed, simple structure net and fast identity speed by the simulation and analysis.At the end, it can process previously the echo signal acquired and extract the character of echo signal in the real environments, use the k-nearest neighbor algorithm and improved RBF neural network to simulate the character extracted. The simulation result indicates that the improved RBF nerve net can identity the real signal acquired effectively, the discrimination can reach 99%.