基于残差网络的钢丝绳损伤图像定量识别

Quantitative Identification of Wire Rope Damage Images Based on Residual Network

目前基于机器视觉的钢丝绳表面损伤检测基本均采用定性检测的方法,在定量检测方面的研究极少,而断丝数量是钢丝绳报废的重要标准,因此,提出一种基于机器视觉和残差网络的钢丝绳表面损伤定量识别方法。将采集到的钢丝绳损伤图像进行批量裁剪,以消去背景噪声;对训练集中的图像利用数据增强技术,进行随机裁剪和随机水平翻转,扩充训练集大小;然后,对数据集中的图像进行归一化和标准化,提高模型的收敛速度;最后将训练集和验证集输入到使用SGD算法优化的残差网络中进行训练,训练结束后再使用测试集对模型进行验证。实验结果表明:经过迭代训练后,模型在测试集上对钢丝绳损伤的定量识别准确率为93.5%。

At present,the surface damage detection of wire rope based on machine vision is mostly qualitative detection,and there is very little research on quantitative detection.The number of broken wires is an important criterion for wire rope scrapped.Therefore,a quantitative identification method for wire rope surface damage based on machine vision and residual network was proposed.The collected wire rope damage images were cropped in batches to remove background noise.The images in the training set were randomly cropped and flipped horizontally by used data enhancement technology to expand the size of the training set.Then,the images in the data set were normalized and standardized to improve the convergence speed of the model.Finally,the training set and the verification set were input into the residual network optimized by using the SGD algorithm for training,and the test set was used to verify the model after the training was completed.The experimental results show that after iterative training,the model s quantitative identification accuracy rate of wire rope damage on the test set is 93.5%.