基于迭代深度学习的缺陷检测

Feature Detection Base on Iterative Deep Learning

随着深度学习的发展,越来越多基于深度学习的应用被推出,深度学习在目标检测,物体识别,语音语义识别等领域都取得了飞跃发展。其中,由于卷积神经网络在图像分类中的广泛应用,现如今的图像识别与传统的图像识别方法已经有了明显的区别。论文使用卷积神经网络对工件缺陷进行检测,针对深度学习在实际应用中出现的小数据集过拟合问题,提出了一种可迭代的深度学习方法来提高识别率并且降低数据的过拟合。

With the development of deep learning,more and more applications based on deep learning are launched.Deeplearning has achieved qualitative development in so many fileds such as object detection,object recognition,speech recognition,se?mantic field.With the widespread use of convolutional neural network in image classification,a marked distinction has occured be?tween the current image recognition and the traditional method of identifying.When we try to find the workpiece defect with the meth?od of t neural network convolution,small data usually result in over-fitting problem.To solve it,we propose a deep learning methodcan be iterative to improve the recognition rate and reduce data over-fitting.