Research On The Recognition And Location Of Multi-target Based On Ultrasonic Phased Array

This paper presents a new solution based on ultrasonic sensor array to solve the issue of shape recognition and location in situations where the optical devices do not operate effectively, such as strong radiation, thick dust, thick fog and so on.According to the review and summarize the research of ultrasonic sensor array in the aspect of shape recognition and location, and combine with the propagation characteristic of ultrasonic sensor in air, this paper designs a kind of two-dimensional ultrasonic testing platform using ultrasonic transducer with a physical dimension of 16 mm and center frequency of 40 k Hz. The platform is composed of ultrasonic sensor array unit, the signal processing unit and embedded heterogeneous unit. The ultrasonic sensor array unit has 8 transmitter transducers and 2×10 receiver transducers, and the transmitter and receiver transducers are interleaved vertically. The signal processing unit is mainly amplifying, filtering, sampling, storage, processing for the weak echo signal. Embedded heterogeneous unit mainly acquire feature extraction of echo signal, realize localization and shape recognition algorithm.In terms of object location, this paper adopts the ultrasonic model of TDOA. To eliminate measuring errors of TOF caused by the multipath reflection, this paper employs cross-correlation algorithm to improve detection precision. Because Fast Fourier transform(FFT) transformation requires a lot of time in cross-correlation algorithm, according to the idea of inter-level serial and outerlevel parallel, the paper introduces CUDA parallel programming technology to accelerate the implementation of the algorithm on GPU.In terms of shape recognition, according to the certain regularity of the echo caused by different shape objects, the paper set up characteristic vector library on time domain method and frequency domain, respectively, and combined with BP neural network to realize the shape recognition. In order to accelerate the realization of BP algorithm on GPU, the paper introduces the parallelism of neuron structure and matrix operation to improve the algorithm.Finally, this paper introduces the CPU serial algorithm and the GPU parallel algorithm to carry out experiments on the ultrasonic platform. According to the experiment, the paper validates the performance of two algorithms, and analyzes the cause of error about object location and shape recognition.