A Real-time Online Aircraft Neural Network System

Abstract

In order to meet the information processing requirements that large amount of heterogeneous input data are in the real-time flight process of aircraft, a neural network is proposed in this paper, including convolution fixed-point sliding IP core, pooling compression quantization IP core and fully connected compression fusion IP core. Heterogeneous sensor data of the aircraft as the input of the system; The recognized result serves as the output of the system. Convolution of sliding window IP core can quickly extract data features by eliminating redundant data sliding window; Pooling compression quantization IP core, using compression quantization technology, improves system execution efficiency; Fully connected compressed fusion IP core is compressed fusion after reduction and quantification, whose output meets the requirements of high reliability and low power consumption of the aircraft online intelligent integration design.