Classification Detection Of Strip Steel Surface Defect Based On Deep Convolution Network

Sheet and strip steel is an important product in the field of iron and steel production.At present,deep convolution network has achieved a good effect in image classification judgment.This paper studies the method of detecting surface defects of strip steel which base on deep convolution network.The specific research content is as follows:(1)For the plate strip surface defect data distribution,the sample size is not balanced and not enough problems.This article adopts the equalization,threshold,image enhancement,noise reduction methods for processing the surface image data and uses the migration learning techniques to a variety of tuning convolution neural network model,in order to select the optimal classification model test categories of surface defect images.(2)Considering the cost problem of computing resources,a VGG16 model base on distributed migration learning was designed.The strategy of momentum gradient and random gradient iterative optimization was adopted to solve the problem.Aiming at the problem of slow convergence caused by BP algorithm to VGG16 feature classifier,a two-point step method was proposed to optimize the update of gradient,which improved the classification performance of VGG16 model.(3)Based on the learned VGG16 model,an online detection system of surface defects of plate and strip steel based on Spark Streaming is implemented.