| 摘要 |
Method for detection of signals with high values of noise relates to measuring engineering and can be used for processing digital signals in acoustic, seismic, electrical, radar systems and systems for transfer, identification and reproduction of digital images. The method is in use for identification of useful signal of artificial neuron networks. One performs training of neuron network with one hidden pass of feeding useful signal without noise to its input and output. After that one analyzes one analyzes values of neuron outputs in hidden layer at feeding to input of neuron network of noised useful signal. The signal is analyzed in parts and number of inputs to neuron network and outputs from it is equal to duration of useful signal divided by pitch of its quantization. If value of neuron outputs in hidden layer is close to unity the signal given is identified as useful one. The method makes it possible to increase area of its application, to increase accuracy of detection of noised signals and to decrease time for signal processing. |
