METHOD OF ALL-ASPECT RECOGNITION OF TYPICAL COMPOSITION OF GROUP AERIAL TARGET FROM CLASS "AIRCRAFT WITH TURBOJET ENGINES" BASED ON KALMAN FILTRATION AND NEURAL NETWORKS
	BOGDANOV A V; IURIN G A; MAKSIMOVICH S V
	2024-07-30
专利权人	RUSSIAN FEDERATION MIN DEFENSE (RUDE-C)
申请日期	2024-07-30
专利号	RU2835772-C1
成果简介	NOVELTY - Invention relates to secondary digital processing of radar signals and can be used to identify a typical composition of a group air target (GAT) from the class "aircraft with turbojet engines (TJE)" at different angles of its radar observation. Disclosed method is based on processing the amplitude-frequency spectrum of the radar signal reflected from the GAT from the "aircraft with turbojet engine" class, spectral components of which are caused by signal reflections from airframes of aircraft of the group and rotating blades of first stages of impellers of low-pressure compressors (LPC) of their engines. Optimal Kalman filtering procedure is calculated with corresponding dynamic models by readings of Doppler frequencies of spectral components of signals, reflected from each airframe of the aircraft of the group and from the rotating blades of the impeller of the first stage of the LPC of each i-th aircraft of the group, having the q-th type, where i=1,…,I; I is the number of aircraft with turbojet engines in the group; q=1,…Q; Q is the number of aircraft types with turbojet engines. In the first neural network preliminary training mode, based on the values of the generated Doppler frequency readings, the Doppler frequency differences are calculated between each Doppler frequency readings, according to which, during K cycles, the first neural network is pre-trained, at the output of which estimates of probabilities of preliminary recognition of the i-th aircraft in the group, having the q-th type, are formed. In the mode of preliminary training of the second neural network in K cycles, values of iq-th autocorrelation functions of Doppler frequency readings are calculated, which are determined by reflection of the radar signal from gliders of aircraft of the group. At the output of the second neural network, estimates of probabilities of preliminary recognition of the i-th aircraft in the group, having the q-th type, are formed. In recognition mode, comparing current view of radar observation of i-th aircraft of q-th type group with threshold value, and, if it is less than the threshold, performing the Doppler frequency readings Kalman filtering, where the iq-th dynamic model is used as the a priori information received during the filtering. As a result, at the output of the i-th Kalman filter, Doppler frequency readings and estimates of their possible differences are generated, which are successively supplied to the inputs of the first neural network, at the outputs of which estimates of the probability of preliminary recognition of the i-th aircraft of the q-th type group are formed, which are compared with a threshold value. If the probabilities are greater than the threshold, the final decision is made that the i-th aircraft of the group is of the q-th type. Second neural network switches to the additional training mode, its w-th inputs receive the generated training samples in the form of values of estimates of the autocorrelation functions of Doppler frequencies. Then, Doppler frequency readings are Kalman filtering, as a result of which estimates are formed at the output of the i-th Kalman filter, from which estimates of values of autocorrelation functions of Doppler frequencies are calculated and they are supplied to w-th inputs of the second neural network. At its outputs, estimates of the probability of preliminary recognition of the i-th aircraft of the q-th type group are formed, and if they are greater than the threshold, the final decision is made that the i-th aircraft of the group has the q-th type. At that, the first neural network switches to the additional training mode, its iq-th inputs receive the generated training samples in the form of a set of Doppler frequency differences, which were obtained on the basis of the Doppler frequency estimate caused by the radar signal reflection from the airframe of each i-th aircraft of the q-th type group and the Doppler frequency readings caused by the signal reflection from the rotating blades of the first stage impeller of the engine LPC of each i-th aircraft of the q-th type. Resultant probability of recognition of each i-th aircraft of the q-th type is determined irrespective of versions of functioning of both neural networks. USE - Data processing. ADVANTAGE - Possibility of recognizing in a pulsed-Doppler radar station with a probability not lower than a given one, the standard composition of the GAT at different angles of its radar observation. 1 cl, 1 dwg, 1 tbl
IPC 分类号	G01S-013/52 ; G01S-013/78 ; G06N-003/02 ; G06T-007/20
国家	俄罗斯
专业领域	信息技术
语种	英语
成果类型	专利
文献类型	科技成果
条目标识符	http://119.78.100.226:8889/handle/3KE4DYBR/15628
专题	中国科学院新疆生态与地理研究所
作者单位	RUSSIAN FEDERATION MIN DEFENSE (RUDE-C)
推荐引用方式 GB/T 7714	BOGDANOV A V,IURIN G A,MAKSIMOVICH S V. METHOD OF ALL-ASPECT RECOGNITION OF TYPICAL COMPOSITION OF GROUP AERIAL TARGET FROM CLASS "AIRCRAFT WITH TURBOJET ENGINES" BASED ON KALMAN FILTRATION AND NEURAL NETWORKS. RU2835772-C1[P]. 2024.

条目包含的文件
条目无相关文件。

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。