基于形状特征张量的集料级配检测研究OA

Existing aggregate gradation detection methods based on machine vision primarily rely on aggregate size information for gradation analysis.However,size information alone does not directly reflect aggregate mass,leading to significant detection errors.To address this issue,a gradation detection method based on shape feature tensors is proposed.In this method,aggregate size and shape features are extracted from two-dimensional images.The size features are used to determine the probability distribution of aggregate size intervals,while shape feature tensors are constructed.A convolutional neural network is then employed to predict aggregate shape factors.By integrating the size interval probability distribution and shape factors,aggregate quantity is converted into equivalent sphere quantity.Finally,equivalent sphere quantity serves as input data for an MCMC algorithm based on Bayesian inference to obtain gradation detection results.Experimental validation shows that the absolute detection error remains within±2.5%,meeting the±5%accuracy requirement for engineering applications and outperforming methods based solely on size features.