摘要
为了提高三维模型渲染时间估计的准确度,采用Spiking神经网络算法进行渲染时间预估。目前,基于Spiking理念的神经网络模型有多种,选择了其中的卷积Spiking神经网络(Convolutional Spike neural network,CSNN)来实现渲染时间计算。首先,建立了基于CSNN的渲染时间预估模型。通过点火时间序列完成编码,从而触发脉冲响应实现数据传递。其次,利用CSNN的权重、卷积核尺寸、偏置等参数来构建菌群优化(Bacterial foraging optimization,BFO)算法,并以渲染时间预估值和实际值的差值作为适应度函数。通过驱化、繁衍和迁徙操作不断更新菌群个体的适应度值来获得最优个体。最后,以最优参数进行CSNN的渲染时间预估。试验结果表明,通过合理设置BFO算法的引力系数、斥力系数和迁徙概率阈值等参数,BFO+CSNN算法能够获得较高的渲染时间预估准确率。相比于其他渲染时间预估算法,BFO+CSNN算法具备更高的渲染时间预估准鲁棒性。
In order to improve the accuracy of rendering time estimation of 3D models,Spiking neural network algorithm is used to estimate the rendering time.At present,there are many kinds of neural network models based on Spiking concept,among which Convolutional Spiking Neural Network(CSNN)is selected to calculate the rendering time.Firstly,a rendering time prediction model based on CSNN is established.The coding is completed by the ignition time sequence,which triggers the impulse response to realize data transmission.Secondly,the Bacterial Foraging Optimization(BFO)algorithm is constructed by using CSNN’s weight,convolution kernel size,offset and other parameters,and the difference between the estimated rendering time and the actual value is used as the fitness function.The best individual can be obtained by constantly updating the fitness value of individual flora through the operations of driving,reproduction and migration.Finally,the rendering time of CSNN is estimated with the optimal parameters.The experimental results show that BFO+CSNN algorithm can achieve high rendering time prediction accuracy by reasonably setting the parameters of BFO algorithm,such as gravity coefficient,repulsion coefficient and migration probability threshold.Compared with other rendering time prediction algorithms,BFO+CSNN algorithm has higher rendering time prediction quasi-stability.
作者
胡博
章毅
蔡柳萍
Hu Bo;Zhang Yi;Cai Liuping(Nanguo Business School,Guangdong University of Foreign Studies,Guangzhou 510440,China;College of Environmental Science and Engineering,Hunan University,Changsha 410082,China;Graduate School,University of Sao Paulo,3500,Tuguegarao,Cagayan Province,Philippines)
出处
《南京理工大学学报》
CAS
CSCD
北大核心
2023年第2期214-220,共7页
Journal of Nanjing University of Science and Technology
基金
国家重点研发计划课题(018YFC0213905)
广东省高校青年创新人才项目(2020KQNCX182)。
关键词
SPIKING神经网络
渲染时间
菌群优化
卷积神经网络
Spike neural network
rendering time
bacterial foraging optimization
convolutional neural network
