Internal Defects Detection Method of the Railway Track Based on Generalization Features Cluster Under Ultrasonic Images

There may be several internal defects in railway track work that have different shapes and distribution rules,and these defects affect the safety of high-speed trains.Establishing reliable detection models and methods for these internal defects remains a challenging task.To address this challenge,in this study,an intelligent detection method based on a generalization feature cluster is proposed for internal defects of railway tracks.First,the defects are classified and counted according to their shape and location features.Then,generalized features of the internal defects are extracted and formulated based on the maximum difference between different types of defects and the maximum tolerance among same defects’types.Finally,the extracted generalized features are expressed by function constraints,and formulated as generalization feature clusters to classify and identify internal defects in the railway track.Furthermore,to improve the detection reliability and speed,a reduced-dimension method of the generalization feature clusters is presented in this paper.Based on this reduced-dimension feature and strongly constrained generalized features,the K-means clustering algorithm is developed for defect clustering,and good clustering results are achieved.Regarding the defects in the rail head region,the clustering accuracy is over 95%,and the Davies-Bouldin index(DBI)index is negligible,which indicates the validation of the proposed generalization features with strong constraints.Experimental results prove that the accuracy of the proposed method based on generalization feature clusters is up to 97.55%,and the average detection time is 0.12 s/frame,which indicates that it performs well in adaptability,high accuracy,and detection speed under complex working environments.The proposed algorithm can effectively detect internal defects in railway tracks using an established generalization feature cluster model.

DSN-BR-Based Online Inspection Method and Application for Surface Defects of Pharmaceutical Products in Aluminum-Plastic Blister Packages

Ensuring high product quality is of paramount importance in pharmaceutical drug manufacturing,as it is subject to rigorous regulatory practices.This study presents a research focused on the development of an on-line detection method and system for identifying surface defects in pharmaceutical products packaged in aluminum-plastic blisters.Firstly,the aluminum-plastic blister packages exhibit multi-scale features and inter-class indistinction.To address this,the deep semantic network with boundary refinement(DSN-BR)model is proposed,which leverages semantic segmentation domain knowledge,to accurately segment the defects in pixel level.Additionally,a specialized image acquisition module that minimizes the impact of ambient light is established,ensuring high-quality image capture.Finally,the image acquisition module,image detection module,and data management module are designed to construct a comprehensive online surface defect detection system.To validate the effectiveness of our approach,we employ a real dataset for instance verification on the implemented system.The experimental results substantiate the outstanding performance of the DSN-BR,achieving the mean intersection over union(MIoU)of 90.5%.Furthermore,the proposed system achieves an inference speed of up to 14.12 f/s,while attaining an F1-Score of 98.25%.These results demonstrate that the system meets the actual needs of the enterprise and provides theoretical and methodological support for intelligent inspection of product surface quality.By standardizing the control process of pharmaceutical manufacturing and improving the management capability of the manufacturing process,our approach holds significant market application prospects.

Algorithmic Scheme for Concurrent Detection and Classification of Printed Circuit Board Defects

An ideal printed circuit board(PCB)defect inspection system can detect defects and classify PCB defect types.Existing defect inspection technologies can identify defects but fail to classify all PCB defect types.This research thus proposes an algorithmic scheme that can detect and categorize all 14-known PCB defect types.In the proposed algorithmic scheme,fuzzy cmeans clustering is used for image segmentation via image subtraction prior to defect detection.Arithmetic and logic operations,the circle hough transform(CHT),morphological reconstruction(MR),and connected component labeling(CCL)are used in defect classification.The algorithmic scheme achieves 100%defect detection and 99.05%defect classification accuracies.The novelty of this research lies in the concurrent use of CHT,MR,and CCL algorithms to accurately detect and classify all 14-known PCB defect types and determine the defect characteristics such as the location,area,and nature of defects.This information is helpful in electronic parts manufacturing for finding the root causes of PCB defects and appropriately adjusting the manufacturing process.Moreover,the algorithmic scheme can be integrated into machine vision to streamline the manufacturing process,improve the PCB quality,and lower the production cost.

Application of improved back-propagation algorithms in classification and detection of scars defects on rails surfaces

An experimental platform with bracket structures,cables,parallel computer and imaging system is designed for defects detecting on steel rails. Meanwhile,an improved gradient descent algorithm based on a self-adaptive learning rate and a fixed momentum factor is developed to train back-propagation neural network for accurate and efficient defects classifications. Detection results of rolling scar defects show that such detection system can achieve accurate positioning to defects edges for its improved noise suppression. More precise characteristic parameters of defects can also be extracted.Furthermore,defects classification is adopted to remedy the limitations of low convergence rate and local minimum. It can also attain the optimal training precision of 0. 00926 with the least 96 iterations. Finally,an enhanced identification rate of 95% has been confirmed for defects by using the detection system. It will also be positive in producing high-quality steel rails and guaranteeing the national transport safety.

Deep Learning-Based Model for Defect Detection and Localization on Photovoltaic Panels

The Problem of Photovoltaic(PV)defects detection and classification has been well studied.Several techniques exist in identifying the defects and localizing them in PV panels that use various features,but suffer to achieve higher performance.An efficient Real-Time Multi Variant Deep learning Model(RMVDM)is presented in this article to handle this issue.The method considers different defects like a spotlight,crack,dust,and micro-cracks to detect the defects as well as loca-lizes the defects.The image data set given has been preprocessed by applying the Region-Based Histogram Approximation(RHA)algorithm.The preprocessed images are applied with Gray Scale Quantization Algorithm(GSQA)to extract the features.Extracted features are trained with a Multi Variant Deep learning model where the model trained with a number of layers belongs to different classes of neurons.Each class neuron has been designed to measure Defect Class Support(DCS).At the test phase,the input image has been applied with different operations,and the features extracted passed through the model trained.The output layer returns a number of DCS values using which the method identifies the class of defect and localizes the defect in the image.Further,the method uses the Higher-Order Texture Localization(HOTL)technique in localizing the defect.The pro-posed model produces efficient results with around 97%in defect detection and localization with higher accuracy and less time complexity.

Description and Classification of Leather Defects Based on Principal Component Analysis

The accurate extraction and classification of leather defects is an important guarantee for the automation and quality evaluation of leather industry. Aiming at the problem of data classification of leather defects,a hierarchical classification for defects is proposed.Firstly,samples are collected according to the method of minimum rectangle,and defects are extracted by image processing method.According to the geometric features of representation, they are divided into dot,line and surface for rough classification. From analysing the data which extracting the defects of geometry,gray and texture,the dominating characteristics can be acquired. Each type of defect by choosing different and representative characteristics,reducing the dimension of the data,and through these characteristics of clustering to achieve convergence effectively,realize extracted accurately,and digitized the defect characteristics,eventually establish the database. The results showthat this method can achieve more than 90% accuracy and greatly improve the accuracy of classification.

Defect Detection in Manufacturing: An Integrated Deep Learning Approach

This research investigates deep learning-based approach for defect detection in the steel production using Severstal steel dataset. The developed system integrates DenseNet121 for classification and DeepLabV3 for segmentation. DenseNet121 achieved high accuracy with defect classification as it achieved 92.34% accuracy during testing. This model significantly outperformed benchmark models like VGG16 and ResNet50, which achieved 72.59% and 92.01% accuracy, respectively. Similarly, for segmentation, DeepLabV3 showed high performance in localizing and categorizing defects, achieving a Dice coefficient of 84.21% during training and 69.77% during validation. The dataset includes steels which have four different types of defects and the DeepLab model was particularly effective with detection of Defect 4, with a Dice coefficient of 87.69% in testing. The model performs suboptimally in segmentation of Defect 1, achieving an accuracy of 64.81%. The overall system’s integration of classification and segmentation, alongside thresholding techniques, resulted in improved precision (92.31%) and reduced false positives. Overall, the proposed deep learning system achieved superior defect detection accuracy and reliability compared to existing models in the literature.

Wood defect detection method with PCA feature fusion and compressed sensing

We used principa/component analysis （PCA） and compressed sensing to detect wood defects from wood plate images. PCA makes it possible to reduce data redundancy and feature dimensions and compressed sensing, used as a elas- sifter, improves identification accuracy. We extracted 25 features, including geometry and regional features, gray-scale texture features, and invariant moment features, from wood board images and then integrated them using PCA, and se- lected eight principal components to express defects. After the fusion process, we used the features to construct a data dic- tionary, and realized the classification of defects by computing the optimal solution of the data dictionary in l1 norm using the least square method. We tested 50 Xylosma samples of live knots, dead knots, and cracks. The average detection time with PCA feature fusion and without were 0.2015 and 0.7125 ms, respectively. The original detection accuracy by SOM neural network was 87 %, but after compressed sensing, it was 92 %.

Soft measurement of wood defects based on LDA feature fusion and compressed sensor images

We proposed a detection method for wood defects based on linear discriminant analysis (LDA) and the use of compressed sensor images. Wood surface images were captured, using a camera Oscar F810C IRF camera, and then the image segmentation was performed, and the defect features were extracted from wood board images. To reduce the processing time, LDA algorithm was used to integrate these features and reduce their dimensions. Features after fusion were used to construct a data dictionary and a compressed sensor was designed to recognize the wood defects types. Of the three major defect types, 50 images live knots, dead knots, and cracks were used to test the effects of this method. The average time for feature fusion and classification was 0.446 ms with the classification accuracy of 94%.

Automated Classification of Snow-Covered Solar Panel Surfaces Based on Deep Learning Approaches

Recently,the demand for renewable energy has increased due to its environmental and economic needs.Solar panels are the mainstay for dealing with solar energy and converting it into another form of usable energy.Solar panels work under suitable climatic conditions that allow the light photons to access the solar cells,as any blocking of sunlight on these cells causes a halt in the panels work and restricts the carry of these photons.Thus,the panels are unable to work under these conditions.A layer of snow forms on the solar panels due to snowfall in areas with low temperatures.Therefore,it causes an insulating layer on solar panels and the inability to produce electrical energy.The detection of snow-covered solar panels is crucial,as it allows us the opportunity to remove snow using some heating techniques more efficiently and restore the photovoltaics system to proper operation.This paper presents five deep learning models,■-16,■-19,ESNET-18,ESNET-50,and ESNET-101,which are used for the recognition and classification of solar panel images.In this paper,two different cases were applied;the first case is performed on the original dataset without trying any kind of preprocessing,and the second case is extreme climate conditions and simulated by generating motion noise.Furthermore,the dataset was replicated using the upsampling technique in order to handle the unbalancing issue.The conducted dataset is divided into three different categories,namely;all_snow,no_snow,and partial snow.The fivemodels are trained,validated,and tested on this dataset under the same conditions 60%training,20%validation,and testing 20%for both cases.The accuracy of the models has been compared and verified to distinguish and classify the processed dataset.The accuracy results in the first case showthat the comparedmodels■-16,■-19,ESNET-18,and ESNET-50 give 0.9592,while ESNET-101 gives 0.9694.In the second case,the models outperformed their counterparts in the first case by evaluating performance,where the accuracy results reached 1.00,0.9545,0.9888,1.00.and 1.00 for■-16,■-19,ESNET-18 and ESNET-50,respectively.Consequently,we conclude that the second case models outperformed their peers.

基于局部特征深度信息的绝缘子小样本缺陷检测

基于深度学习的目标检测技术已广泛应用于绝缘子缺陷检测中,然而现有目标检测算法主要基于大量缺陷样本训练网络模型,无法对少样本缺陷进行准确识别。针对绝缘子缺陷检测过程中缺陷样本量不足的问题,该文提出了一种基于局部特征深度信息的绝缘子小样本缺陷检测方法。首先采用旋转目标检测网络改进Faster R-CNN(faster region-based convolutional neural network)模型提取绝缘子串区域,然后对绝缘子串特征进行划分,提取绝缘子串局部特征并基于深度推土距离(deep earth mover’s distance,Deep EMD)网络实现小样本缺陷检测。实验结果表明,在玻璃绝缘子自爆缺陷检测中,所提出方法采用2张训练样本可取得与现有目标检测方法 200张训练样本相同的效果,采用10张训练样本的绝缘子自爆检测在与真值框的交并比阈值为0.5至0.95之间的平均精度(mean average precision,mAP)达到0.65,该方法为小样本电力设备缺陷智能化检测提供了新的方法和思路。

基于改进YOLOv5-LITE轻量级的配电组件缺陷识别

为对配电组件缺陷进行精确快速的定位和识别,提出一种基于改进YOLOv5-LITE轻量级的配电组件缺陷识别方法。为使模型便于部署至移动设备终端,该方法使用ShuffleNetV2作为骨干网提取特征构建YOLOv5-LITE轻量级神经网络模型,并摘除ShuffleNetV2的1024卷积和5×5池化,采用全局平均池化操作替代,降低网络参数量,提升模型检测速度;通过引入有利于细粒度目标检测的152×152特征层,实现了对大、中、小尺度的缺陷检测;在PANet架构中采用深度可分离卷积代替下采样使得网络更加轻量化。实验结果表明:该方法能够识别电缆脱离垫片、电缆与绝缘子脱落、无环绝缘子3种缺陷,其检测精度分别达到92%、95%、95%,网络参数量约为YOLOv5的1/4,检测速度达到2 ms/张。所提出的方法具有实时性、准确率高、轻量化等特点。

局部和全局特征融合的太阳能电池片表面缺陷检测

太阳能电池片表面缺陷具有类内差异大、类间差异小和背景特征复杂等特点,因此,要实现高精度的太阳能电池片表面缺陷自动检测是一项富有挑战性的任务。针对此问题,该文提出融合局部和全局特征的卷积视觉Transformer网络(CViT-Net),首先采用Ghost聚焦(G-C2F)模块提取电池片缺陷局部特征;然后引进坐标注意力强调缺陷特征并抑制背景特征;最后构建Ghost视觉(G-ViT)模块融合电池片缺陷局部特征和全局特征。同时,针对不同检测精度和模型参数量,分别提供了CViT-Net-S和CViT-Net-L两种网络结构。实验结果表明,与经典MobileVit、MobileNetV3和GhostNet轻量级网络相比,CViT-Net-S对电池片分类准确率分别提升了1.4%、2.3%和1.3%,对电池片检测mAP50分别提升了2.7%、0.3%和0.8%;与ResNet50、RegNet网络相比,CViT-Net-L分类准确率分别提升了0.72%和0.7%,检测mAP50分别提升了3.9%、1.3%;与先进YOLOv6、YOLOv7和YOLOv8检测网络相比,作为骨干网络的CViT-Net-S、CViT-Net-L结构在mAP和mAP50指标上仍保持良好检测效果。结果证明本文算法在太阳能电池片表面缺陷检测领域具有应用价值。

AI大模型赋能网络流量分类概述

提出一个通用的AI驱动的网络流量分类框架,阐述了所涉及的工作流程、分类目标、设计原则以及典型场景等,并提出了一个基于BERT的网络流量分类模型,通过将输入的分组净荷进行向量化嵌入,然后送入BERT进行预训练,用于实现流量数据的上下文理解并捕获双向特征,然后对接一个全连接网络对分类下游任务进行微调,从而实现流量分类。通过与AE、VAE、ByteSGAN 3个经典的流量分类深度学习模型在CICIDS2017公开数据集上进行对比,发现BERT的精度明显高于其他方法。

基于人工智能的皮革材料缺陷检测与识别方法研究

本文在分析现代皮革材料缺陷检测与识别方法的基础上,尝试将人工智能技术融入该类材料的检测环节,构建了一种基于深度学习无监督+二分类需求的皮革材料缺陷检测智能化系统,并对该系统工作时的缺陷检测效率、误检率、缺陷识别情况等进行测试。结果表明,该方法以深度学习为核心,可以获得较高的皮革材料缺陷检测效率、较低的误检率和较好的自主学习能力。

火炮摇架焊缝缺陷智能分类

针对火炮摇架结构复杂,焊缝内部缺陷检测效果不理想,选择对非平面物体和复杂结构体均适用的超声相控阵检测技术对摇架焊缝缺陷进行检测。将得到的超声相控阵图谱与ResNeXt网络模型相结合,实现焊缝缺陷的智能分类。将SK卷积单元引入ResNeXt网络模型,对摇架焊缝缺陷进行定性分析。改进后的网络模型比原ResNeXt网络的分类准确率提升5.5%,最终达到98.2%。

基于深度智能视觉的表面缺陷检测研究进展

基于深度智能视觉的表面缺陷检测研究在制造业中起着越发重要的作用,本文阐述深度智能视觉的表面缺陷检测在现代工业质检中的重要性,对现有研究进展进行梳理总结。深度智能视觉以机器视觉和深度学习为技术基础,为不同工业场景提供高精高效的表面缺陷检测算法。本文从检测细粒度的角度将表面缺陷检测分为表面缺陷分类、定位、分割检测3个部分,并分别对分类、定位、分割方法进行系统综述,梳理现有表面缺陷检测研究的问题和思路。分类检测针对数据和缺陷图形特征问题进行研究,因其基础性和易拓展性于不同工业场景的应用呈现分散发展;定位检测以模型框架、矩形框检测和标注成本为主要问题,表现出追求轻量化和特征融合机制的研究趋势;分割检测更关注图像细节特征。通过研究分类、定位、分割的多任务模型框架以探索分类、分割检测之间的互补性。最后总结目前表面缺陷检测研究存在的问题,并对发展趋势进行展望。

基于机器视觉的钢轨表面面型缺陷分类实验设计

随着城市轨道交通的飞速发展,实现钢轨表面缺陷实时检测对铁路行业稳步发展意义重大。如何实时检测钢轨表面缺陷是保障铁路运行安全亟须解决的一个关键问题。鉴于此,设计了一套基于机器视觉的钢轨表面缺陷检测实验仿真方法。搭建图像采集、图像预处理和缺陷分类等模块;提出自拟合亮度调整算法完成像素值统计,得到清晰的缺陷特征图像;用750组数据训练网络权值,实现缺陷分类预测;经过数据分析和误差评估,识别准确率在90%以上,相关系数高达0.96,单幅图像平均耗时1.267 s,测试表明,所提方法能准确、高效地实现钢轨表面缺陷信息的缺陷分类与识别。

磁信号检测高温合金内外壁缺陷分类研究

为区分高温合金内外壁缺陷,采用高精度弱磁传感器对高温合金人工槽型缺陷进行检测。从有缺陷的一面检测为正面(即外壁缺陷),从另一面检测为反面(相当于内壁缺陷),发现单侧表面缺陷检测信号出现极性相反的情况,无法通过极性相反这一特征区分内外壁缺陷。针对这一现象,基于支持向量机中的分类功能,对采集到的弱磁信号进行特征提取获得面积、幅值和占宽3种特征量建立分类数据库,进行内外壁缺陷分类识别,并采用不同的核函数建立分类模型进行比较,最终发现径向基核函数分类模型正确率最高,为84.37%,且经过样本库外的试件验证后正确率仍有81.25%。结果表明该算法能够有效地对高温合金缺陷信号进行分析和辨识,具有一定的实用价值。

可区分应力集中和缺陷的双线圈共磁芯式梯度测磁传感装置设计

针对复杂构件特殊位置熔焊缝无法全覆盖检测的工程问题,基于磁检测技术,设计了一种双线圈共磁芯式梯度测磁传感装置。以S06马氏体不锈钢加工的熔焊缝试块为实验对象,通过对比不同提离高度下焊接缺陷和应力集中处的磁信号检测结果,以及对传感装置的性能测试,确定装置参数。结果表明,在应力集中和焊接缺陷处,磁感应强度均随着传感器提离值的增大而减小,且在一定的提离高度下,缺陷磁异常特征消失,而应力集中处磁异常信号仍然存在,可据此来区分应力集中和焊接缺陷。射线检测结果验证了所设计传感装置在焊缝缺陷检测上的可行性与有效性。对于材料磁性和焊接工艺相似的熔焊缝构件,可采取同样的试验方法标定装置参数,以扩大所设计装置的适用性。