FIELD: hydro acoustics.
SUBSTANCE: invention relates to hydroacoustics and can be used to construct a marine monitoring system operated by a programmable neural network. Disclosed method is realized based on computational operations of artificial neural networks, means of marine instrument-making and long-range parametric reception of waves in the sound and infrasound frequency bands. Application of computer operations of artificial neural networks with preliminary compression of information and replenished libraries of mathematically processed images of spectrograms of objects speeds up the recognition process and increases probability of classification as surface, and underwater targets detected during far parametric reception. Method of forming a marine monitoring system with a programmable neuron network control system, comprising: first, in a marine environment, forming a zone of nonlinear interaction and parametric conversion of pumping waves by object waves by emitting a low-frequency acoustic signal therein, for this purpose, radiator and receiving transducer are arranged at opposite boundaries of controlled section of marine environment. Pumping waves, modulated by object waves, are received and amplified in parametric conversion band, their frequency-time scale is transferred to high-frequency region, narrow-band spectral analysis is carried out, parametric components of the total or difference frequency are selected, from which, based on the time and parametric conversion of waves, the characteristics of the object signals are restored. Principal difference from prototype is that amplitude-frequency characteristics of object signals obtained using narrow-band spectral analysis in the reception channel, processing and recording of signals, are supplied to additionally introduced neuron network control system to preprocessing unit filter input. Filtration of discrete components is carried out in accordance with frequency range specified for neural network, then thinned set of discrete components is supplied to input of data preparation unit preliminary processing unit, where statistical parameters of object are calculated and compressed by Kolmogorov-Khinchin method. Next to the first input of the target type categorizer of the neuronet recognition module, data are output from the preprocessing unit data preparation unit output, and its second input receives data from neuronet recognition module learning unit, in memory of which mathematically processed images of sea target spectrograms are recorded. Artificial neuron network of target type categorizer is then tuned by classification criteria of targets, computer operations are started, according to the results of which its weight coefficients are corrected and a conclusion on degree of belonging of the analysed spectral region to the classification object (surface or underwater object) is formed. Further, the signal is transmitted to the neural network control unit control unit, where commands are generated and sent to the radiation path to the input of the stabilized frequency pumping signal generator, where pumping signals of the sea medium are generated in accordance with the tasks and conditions of long-term marine monitoring.
EFFECT: providing long-range parametric reception of waves in the sound and infrasound frequency bands with correction of the process of generating emitted pumping signals of the medium in accordance with the tasks and conditions of long-term marine monitoring, as well as recognition (classification) of detected objects based on computer operations of artificial neural networks using preliminary information compression on object and replenished libraries of mathematically processed images of spectrograms of targets.
1 cl, 8 dwg
