METHOD OF RECOGNIZING TYPE OF AERIAL TARGET FROM CLASS "AIRCRAFT WITH TURBOJET ENGINE" BASED ON NEURAL NETWORK Russian patent published in 2024 - IPC G01S13/52 G01S7/41 G06N3/02 G06T7/20 

FIELD: radar.

SUBSTANCE: invention relates to radar and can be used to identify in pulse Doppler radar station various types of air target from class "Aircraft with turbojet engine (TJE)" under stationary conditions of its flight based on neural network. In the disclosed method, the radar signal S(t) reflected from the aircraft with the turbojet engine is subjected to narrow-band Doppler filtering in the Fast Fourier transform (FFT) unit and converted into an amplitude-frequency spectrum (AFS) S(f)={F,A}, where F – values of frequencies of spectral components, A – corresponding amplitudes. In the Spectrum truncation and normalization device, a sub-range of finding informative features is calculated and the AFS is truncated by the calculated sub-range and normalized by amplitude. Truncated and normalized array of amplitudes A is supplied to the corresponding inputs of the Estimating device, which is a set W of pre-trained neural networks in the Neural network model unit. Each neural network performs recognition of only one of W types of a priori class alphabet. Output of the w-th neural network in the Estimating device will be the probability P₁ of coincidence of the received image of normalized amplitudes A with one of the W types of the a priori class alphabet and probability P₂ of mismatch of the supplied image of normalized amplitudes A formed by the neural network of the generalized image. From the output of the Estimating device, the vector of estimations of the probabilities of referring the image of normalized amplitudes to the w type of the air target from W possible arrives to the input of the Resolving device. In the Resolving device, the probabilities P₁ and P₂ are compared with the given threshold probabilities P_tr1 and P_tr2respectively. If the condition P_1w>P_tr1∪P_2w>P_tr2, based on the criterion of maximum probability P_1w, an unambiguous assignment is made to a specific type w of an aerial target from the a priori class alphabet, the value of which is transmitted to the output of the Resolving device as a recognition result. Training of neural networks is carried out in a Model of a neural network on types W of the a priori alphabet of classes using a model reflected radar signal.

EFFECT: development of a method which enables to recognize in a pulse-Doppler radar station with a probability not lower than a given one, the type of an aerial target from the class "Aircraft with turbojet engine" under stationary conditions of its flight.

1 cl, 1 dwg, 1 tbl