An image feature description method based on impulse neural network

Abstract

The invention discloses an image feature description method based on pulse neural network, which comprises the following steps: S1, preprocessing the original image; S2, the convolution layer of the Gabor filter of the impulse neural network is thinned by using the Sobel operator; 3, inputting the processed image into a sparse Gabor filt convolution layer for feature extraction; S4, realizing feature extraction; S5, spiking encoding and sampling processing are carried out by using the encoding layer; S6, according to the pulse sequence, using the improved Tempotron algorithm to learn the connection weights from the coding layer neurons to the learning layer neurons; S7, describing the processed image and describing the image features. The invention solves the problems existing in the priorart that the feature selection and extraction are tedious, the calculation is complex, the extracted feature is weak in distinguishing the object, the important feature is easy to lose and change, and the image can not be completely restored according to the memory.

Description

technical field [0001] The invention belongs to the field of intelligent computing, and in particular relates to an image feature description method based on a pulse neural network. Background technique [0002] In terms of the ability to recognize objects, the biological visual system has a stronger ability than the current computer vision system. When two different images appear to the left and right of the human experimenter, the experimenter can select the target object within 120-130 milliseconds. If we assume that the processing latency in the human visual system is 20-30ms, this means that the underlying visual processing can be done in 100ms or less. In the latest EEG experiments in monkeys, data recorded from the temporal lobe cortex (IT) revealed spike firing (Spike) over a time period of 12.5 ms and a signal for a visual stimulus generated only about 100 ms after stimulus initiation. [0003] Computer vision is the use of computers and related equipment to simul...

AI Technical Summary

Problems solved by technology

[0006] Aiming at the above-mentioned deficiencies in the prior art, the present invention provides an image feature description method based on a spiking neural network, which is simple in feature selection and extraction, convenient in calculation, strong in distinguishability of extracted features, good in extraction effect and high in practicability, ensuring The invariance of extracted features solves the problems of cumbersome feature selection and extraction in the prior art, complex calculations, weak distinguishability of extracted features for objects, easy loss and change of important features, and the inability to restore images based on memory. question

Images