基于深度学习的缺陷识别算法设计

Design of Defect Recognition Algorithms Based on Deep Learning

本文对基于深度学习的缺陷识别算法进行了研究,以轮毂缺陷为例,提出了一种基于改进快速区域卷积神经网络(Faster Region Convolutional Neural Networks,Faster R-CNN)的轮毂缺陷识别算法。测试结果表明相比基于原始Faster R-CNN的轮毂缺陷识别算法,本文算法对轮毂缺陷进行识别的精确度提高了3%,维持在98%以上,每一秒识别图像增加了4帧以上,稳定在每秒识别15帧以上,改进方法具有可行性和优越性。

This article studies defect recognition algorithms based on deep learning,taking wheel hub defects as an example,and proposes a wheel hub defect recognition algorithm based on an improved Faster Region Convolutional Neural Network(Faster R-CNN).The test results show that compared with the wheel hub defect recognition algorithm based on the original Faster R-CNN,the accuracy of the proposed algorithm in identifying wheel hub defects has been improved by 3%,maintained at over 98%,and the recognition image has increased by more than 4 frames per second,stabilizing at more than 15 frames per second.The improved method is feasible and superior.