Research On Automatic Obstacle Avoidance System Of Intelligent Vehicle Based On Spike Neural Network

With the rapid development of science and technology,people are quickly entering the era of artificial intelligence.Smart cities are currently a hot research topic.Smart vehicles as part of smart cities are also one of the key research objects in the current era.Due to the substantial improvement of people’s living standards,the utilization rate of vehicles has also soared,but the problem of road traffic safety has become more and more serious,and it has also caused environmental pollution.How to solve these problems has become the focus of increasing society’s attention.To solve these problems,it is necessary to note on how to improve ability of the vehicle’s automatic obstacle avoidance and prediction,sensitivity to changes in the surrounding environment.In this paper,related research and design are carried out for the obstacle avoidance ability of the vehicle,including the design of obstacle avoidance control module and the analysis of the algorithm based on SNN.This article first briefly describes the traditional obstacle avoidance model algorithms,such as APF and VFF,as well as fuzzy control algorithms.At the same time,according to the obstacle avoidance principle of the algorithm,these algorithms are analyzed for the obstacle avoidance behavior problem of the car.The shortcomings of the above,and then lead to the obstacle avoidance control algorithm of the trolley based on the pulse neural network.Secondly,explaining the biological significance and the basic structure of the SNN,the dynamic principle and basic application of the spike neuron model are briefly introduced,and the neural network model are described from two aspects of the coding mechanism and the network training algorithm.Then it introduces the development process of STDP algorithm in detail,and improves the shortcomings of the fully connected network based on STDP rules.The way to improve is to dynamically change the network probability connection and add a competitive algorithm mechanism.It is verified by corresponding simulation experiments that the improved network model reduces the training time of the network to a certain extent.The data input into the neural network needs to be encoded,and the encoded pulse sequence can be recognized by the network.In this paper,the Poisson coding mechanism is improved by adding a scale factor and a bias coefficient to it.Through experiments,it is known that the improved Poisson coding mechanism can dynamically adjust the number of pulses sent by the impulse neuron to increase it.Classification accuracy of network training.Finally,the corresponding simulations are carried out on different network models.The experiment shows that the pulse neural network can better realize the obstacle avoidance behavior of the car compared with other neural networks.Based on the impulse neural network,four different network structure models are designed from the perspective of changing the network connection probability and increasing the algorithm mechanism,and simulation experiments are performed on them respectively.The experiments show that the impulse neural network based on probabilistic connection competition learning Compared with the fully connected impulse neural network based on STDP rules,the accuracy rate is increased by 1.6%.Changing the network connection structure and increasing the competition algorithm mechanism can reduce the network calculation rate to the greatest extent and shorten the training time.And under certain conditions,the accuracy of the impulse neural network based on probabilistic connection competition learning can be increased by 2.6%,and the accuracy of obstacle avoidance classification has reached the best,which further shows that the algorithm model selected in this paper has certain research value.