Edge detection method based on long-short-term synaptic complementary neural network

Abstract

The invention relates to an edge detection method based on a long-short-term synaptic complementary neural network. A neuron network with a long-short-term synaptic complementary characteristic is constructed, and comprises a color antagonism weighted coding module, a discharge time coding module and a long-short-term synaptic complementary coding module. In the color antagonism weighted coding module, weighted coding is performed on a color antagonism channel of a to-be-detected image; in the discharge time coding module, discharge time coding of the weighted coding response is realized; in the long-short-term synaptic complementary coding module, long-short-term synaptic plasticity coding is realized based on neuron group discharge activity space-time dependence and synchronous dischargecharacteristics, long-short-term synaptic result complementary fusion is realized, and an edge response is obtained by coding a time information flow; and a final edge result is obtained through normalization and gray scale mapping processing. According to the method, the complementary effect of the plasticity of long-short-term synaptic in the edge detection process is considered, and a good detection effect is achieved on images with complex backgrounds and many weak edges.

Description

Technical field [0001] The invention belongs to the field of visual nerve computing, and mainly relates to an edge detection method based on a synapse complementary neuron network with a long and short duration. Background technique [0002] Image edge detection technology is the basis of image segmentation and target area recognition technologies. Traditional edge detection methods often use gradient operators such as Sobel operator to measure the abrupt step of the image edge, but the edge location is not accurate enough in complex background conditions; there are also detection methods based on LOG filter, which are used to achieve edge location. At the same time, the ability to eliminate noise is weak. Considering that in the biological visual system, the mechanism of synaptic plasticity is important for realizing the visual perception of the brain. For example, there are studies that simulate synaptic plasticity from the perspective of receptive field deformation, but they ...

AI Technical Summary

Problems solved by technology

Traditional edge detection methods often use gradient operators such as the Sobel operator to measure the abrupt step of the edge of the image, but the positioning of the edge is not accurate enough in the case of a complex background; there are also detection methods based on the LOG filter. At the same time, the ability to eliminate noise is weak

Considering that in the biological visual system, the mechanism of synaptic plasticity is very important for realizing the visual perception function of the brain. For example, some studies have simulated synaptic plasticity from the perspective of receptive field deformation, but failed to study synaptic plasticity in the process of image information flow encoding. Dynamic regulation of neuron firing; there are also studies focusing on a single synaptic plasticity, for example, methods based on long-term synaptic plasticity only consider synaptic connections in a long-term scale, which will lead to weak signals in the process of information transmission Not sensitive enough to change to effectively capture weak edges in the image and ignores presynaptic correlations that may lead to fragmentation of image edges

For example, there are also methods based on short-term synaptic plasticity, but they ignore the synaptic dynamic adjustment in the long-term range, which is greatly disturbed by noise.

Images