Parallelization Processing Of Deep Learning Algorithm And Its Application

With the rapid development of high-speed trains, high-speed train security issues gradually attracted people’s attention. High-speed train vibration monitoring data for the analysis of the performance of the train service provided the conditions. However, how timely and accurate fault mining properties from these massive data for fault diagnosis, given that existing problems. Likewise, the modern battlefield radar countermeasures facing increasingly complex electromagnetic environment issues. How to extract from a variety of complex, long-term accumulation of reconnaissance data in a new and effective features of radar emitter signal sorting identification is difficult.Deeping learning in recent years in the field of machine learning research focus, Deep Belief Net (DBN) as a pioneer in constructing such a deep structure, with a strong ability to express the data features. Therefore, DBN algorithm is applied separately to the high-speed train fault diagnosis and identification of radar emitter signal sorting them.Firstly, we analyzed the Deep Belief Net algorithm, the DBN algorithm combined with Hadoop platform constructed DBN algorithm parallelization. And selection criteria MNIST dataset experimental results show overall digital recognition rate of 98% speedup increased to 3, which illustrate parallelism DBN algorithm has good performance results in the recognition accuracy, parallelization efficiency.Then the high-speed train normal and fault analysis of vibration data in time domain, frequency domain characteristics. After the use of the spectrum coefficient signal initialization visible cell DBN, using parallelism DBN algorithm for high-speed train vibration data for deep fault feature extraction and classification. Experimental results show that the statistical results of the better channel, the train fault identification recognition rate of 96%,89% identify the fault location, and the efficiency of the algorithm has some improvement.Lastly, five-parameter analysis of radar emitter signal feature five kinds System. Visible cells use five parameters characteristic of the signal initialization DBN will parallelize DBN algorithm to identify radar emitter signal sorting, the realization of radar emitter signals abstract feature extraction and classification. Experimental results show that the radar emitter signals average correct classification rate of 94%, and has improved the speed of classification.