基于堆叠降噪自动编码器的胶囊缺陷检测方法

目前医用胶囊生产过程中的缺陷检测主要由人工完成,费时费力,容易受主观因素的影响。提出一种基于堆叠降噪自动编码器的胶囊表面缺陷检测方法,该方法首先建立深度自动编码器网络,并根据缺陷样本进行降噪训练,获取网络的初始权值；然后通过BP算法进行微调,得到训练样本到无缺陷模板之间的映射关系；最后利用重构图像与缺陷图像之间的对比关系,实现测试样本的缺陷检测。实验表明,堆叠降噪自动编码器较好地建立了上述映射关系,能快速、准确地进行缺陷检测,对噪声具有很强的鲁棒性和稳定性。

Capsule Defects Detection Based on Stacked Denoising Autoencoders

At present defects of capsules are detected mainly by manual operation,which is time-consuming and needs high labor costs,besides,it is easily misled by subjective factors.This paper proposed a method of detection of capsules surface defects based on stacked denoising autoencoders (SDAE).Our method firstly establishes deep autoencoders networks and trains using a denoising criterion according to the defect samples to obtain the initial weights at first.Then,BP algorithm fine-tunes the network parameters to get the mapping relationship between the training sample and defect-free template.Finally, defect detection of the testing samples is finished by comparing the reconstruction image and defect image.Experimental results show that SDAE perfectly establishes the mapping relationship,which is robust and stable to noise,and can quickly detect defects with high accuracy.