摘要
为了实现复杂条件下配电箱金属表面腐蚀等级的快速、较准确检测,结合深度学习对金属腐蚀检测进行深入研究。现场采集了湖北电力公司中配电箱金属表面腐蚀图片,且对配电箱所属区域环境进行了较详细的分析,获得了较好的金属腐蚀等级标签。在使用连续多层小型卷积滤波器的基础上,添加SENet特征提取模块,提出MS1Net卷积神经网络模型,并使用交叉熵损失函数对MS1Net进行优化。为了验证MS1Net有效性,针对同一网络不同损失函数之间进行对比实验,结果验证交叉熵损失函数收敛更快,loss最低值达到0.077 0。针对多个网络结构如ZFNet、VGG16和MS1Net进行对比实验,最终表明MS1Net能够更快速、更准确地对金属表面腐蚀等级进行检测,且检测准确率为98.44%。
In order to improve the speed and accuracy of metal corrosion detection of the distribution box's metal surface,we use deep learning to conduct in-depth research on metal corrosion detection.We collected on-site pictures of the metal surface corrosion in Hubei Electric Power Company.Through the detailed analysis of the environment where the distribution box belonged,we marked each picture with a sufficiently accurate metal corrosion grade label.On the basis of using continuous multilayer small convolution filter,we added the SENet feature extraction module,proposed the MS1Net convolutional neural network model,and used the cross-entropy loss function to optimize MS1Net.In order to verify the effectiveness of MS1Net,we compared some different loss functions under the same network structure.The experimental results show that the cross-entropy loss function converges faster and the lowest loss value reaches 0.0770.Furthermore,we compared several network structures such as ZFNet,VGG16 with MS1Net.The results show that the MS1Net can detect the corrosion grade of the metal surface more quickly and accurately,and the detection accuracy is 98.44%.
作者
谭暑秋
石林
张建勋
Tan Shuqiu;Shi Lin;Zhang Jianxun(College of Computer Science and Engineering,Chongqing University of Technology,Chongqing 400054,China)
出处
《计算机应用与软件》
北大核心
2023年第9期229-235,共7页
Computer Applications and Software
基金
重庆市基础研究与前沿探索项目(cstc2018jcyjAX0287)。
