基于轮胎X光图像样本重采样图像缺陷检测

针对生成对抗网络轮胎X光图像缺陷检测,训练阶段生成器会丢失部分图像特征,并且难以确定样本的潜在空间维度,导致部分不必要的图像特征重建。为了解决这些问题,构建了样本重采样生成对抗网络SRGAN(Sample Resampling Generate Adversarial Networks),生成器以VQ-VAE为基本框架,利用注意力特征融合模块(Atten-tion Feature Fusion,AFF)搭建了新的跳连层,并在SRGAN的生成器中加入了转换损失函数LVQ。最后,使用自制的轮胎X光图像数据集对SRGAN和已经提出的部分生成对抗网络模型进行训练和测试,并将得到的AUC值进行对比,进一步证明了SRGAN具有更好的图像缺陷检测能力。

基于梯度引导加权‒延迟负梯度衰减损失的长尾图像缺陷检测

针对目前图像缺陷检测模型对长尾缺陷数据集中尾部类检测效果较差的问题,提出一个基于梯度引导加权‒延迟负梯度衰减损失(GGW-DND Loss)。首先,根据检测器分类节点的累积梯度比值分别对正负梯度重新加权,减轻尾部类分类器的受抑制状态;其次,当模型优化到一定阶段时,直接降低每个节点产生的负梯度,以增强尾部类分类器的泛化能力。实验结果表明,在自制图像缺陷数据集和NEU-DET(NEU surface defect database for Defect Detection Task)上,所提损失的尾部类平均精度均值(mAP)优于二分类交叉熵损失(BCE Loss),分别提高了32.02和7.40个百分点;与EQL v2(EQualization Loss v2)相比,分别提高了2.20和0.82个百分点,验证了所提损失能有效提升网络对尾部类的检测性能。

基于FAMGAN的轮胎X光图像缺陷检测

针对气泡缺陷特征和图像背景像素差异较小、检测困难的问题,以Skip-GANomaly为基础框架,提出了融合注意力机制生成对抗网络(FAMGAN),首先,生成器中编码器和解码器之间的跳连层由注意力特征融合模块(AFF)和注意力机制模块(CBAM)构成,提高了对目标特征的关注、减少了图像特征丢失;然后,在判别器中加入联合上采样模块(JPU),提高了模型检测图像缺陷的速度。最后,将本文提出的FAMGAN网络与近几年经典的生成对抗网络在自制的轮胎缺陷数据集上进行训练、测试和评估。实验结果表明,本文提出的网络对轮胎气泡缺陷检测的精度达到0.837,相比于Skip-GANomaly网络提高了近30%。

改进YOLOv5n的管道DR缺陷图像检测方法

近年来,数字射线成像技术(digital radiography,DR)由于其独有优势已被广泛应用于工业管道无损检测。为提高管道DR缺陷图像检测精度,提出了一种改进的YOLOv5n管道DR缺陷图像检测方法。该方法有两点贡献,针对目标检测网络中分类和回归两个任务存在冲突的问题,设计了任务独立解耦检测头,通过分别为两类任务构建独立的特征图实现解耦。为了缓解解耦检测头模块带来的参数量增加问题,引入了轻量化的深度可分离卷积替代标准卷积,在保证精度的同时,减少模型参数量。实验结果表明,在管道缺陷数据集上,该方法的mAP@0.5比YOLOv5n提高0.9个百分点。与YOLOv4、Faster-RCNN和SSD等其他几种目标检测模型的对比实验表明,该方法在mAP@0.5、参数量和计算量上都达到最优,有效提高了管道DR缺陷图像检测的性能。

基于改进的CenterNet变电站设备红外温度检测方法

红外检测能够检测变电站电力设备温度异常,降低安全事故发生的概率,因此,提出一种基于改进的CenterNet目标检测算法模型CenterNet_PRO;该算法采用了ShuffleNet V1/V2作为骨干网络、引入了FPN来提取多尺度特征,为了克服不同尺度目标检测的难点、增加旋转角度回归分支,用于预测目标的旋转角度以及改进的IoU Loss进行优化,进一步提高模型检测速度和准确率;通过阈值分割法提取电力设备表面温度并分析计算,设计制定电力设备温度缺陷判断规范、温度警告阈值,根据该规范即可判断电力设备的相关缺陷;实验结果表明,改进的CenterNet模型平均精度达到了90%,相比于传统的CenterNet模型,平均精度提高了1.3个百分点,可以满足实际变电站场景下对电力设备红外检测的高要求。

面向铝合金焊缝DR图像缺陷的Sim-YOLOv8目标检测模型

针对当前目标检测算法在铝合金焊缝数字射线成像(DR)图像缺陷检测中精度不足的问题,提出了一种基于YOLOv8的改进模型Sim-YOLOv8。首先改进C2f,通过增加SimAM模块提升模型的整体性能;其次,针对部分像素较小的气孔和夹渣缺陷,将首层卷积模块更换为Focus模块,以提升模型对小目标的检测能力;最后添加WIoU损失函数,以提高模型锚框的质量,从而提高检测效果。实验结果表明:在阈值为0.5的前提下,Sim-YOLOv8模型对气孔、夹杂、未焊透这三类缺陷检测的平均精度(mAP@0.5)达到了93.6%、94.4%、97.3%,较原模型分别提高了2.5、1.9和1.7个百分点,具有更好的焊缝缺陷检测效果。

基于改进型PCNN图像分割技术的金属表面检测

对金属表面缺陷中的轧入氧化皮、斑点和划痕3种缺陷检测设计了一种有效方法。首先将目标图片进行灰度值归一化增强图像,提高其视觉效果方便后续图像处理。然后应用改进型脉冲耦合神经网络图像分割技术将目标图片分割得到二值图像,因为三种缺陷各有其独有的特征,据此将二值图像进行分析对比,最后对已经检测出的三种缺陷采取不同的方法标注缺陷位置。在分割方法上与Gabor分割方法进行对比实验,实验结果表明所设计的方法对三种缺陷的检测是十分迅捷且有效的,是可以用于现代化工厂缺陷检测的。

人工智能在非物质文化遗产数字化保护中的应用

在科学技术变革的冲击下,人工智能逐渐介入非物质文化遗产数字化保护与传承工作中。从信息技术在非遗保护传承中的运用角度,对相关文献的研究进展进行了分析和梳理。融入人工智能的相关概念和理论,整理多个应用人工智能进行非遗保护的案例,围绕遗产图像缺陷检测、遗产图像与建筑识别、语义搜索与知识推荐、音视频分析整理、遗产虚拟复原等方面进行具体的阐述。最后介绍了目前人工智能技术应用于非遗数字化保护的优势和存在的问题,为同类研究和实践工作提供了借鉴和参考。

A method for workpiece surface small-defect detection based on CutMix and YOLOv3

Surface small defects are often missed and incorrectly detected due to their small quantity and unapparent visual features.A method named CSYOLOv3,which is based on CutMix and YOLOv3,is proposed to solve such a problem.First,a four-image CutMix method is used to increase the small-defect quantity,and the process is dynamically adjusted based on the beta distribution.Then,the classic YOLOv3 is improved to detect small defects accurately.The shallow and large feature maps are split,and several of them are merged with the feature maps of the predicted branch to preserve the shallow features.The loss function of YOLOv3 is optimized and weighted to improve the attention to small defects.Finally,this method is used to detect 512×512 pixel images under RTX 2060Ti GPU,which can reach the speed of 14.09 frame/s,and the mAP is 71.80%,which is 5%-10%higher than that of other methods.For small defects below 64×64 pixels,the mAP of the method reaches 64.15%,which is 14%higher than that of YOLOv3-GIoU.The surface defects of the workpiece can be effectively detected by the proposed method,and the performance in detecting small defects is significantly improved.

Automatic Defect Detection and Grading of Single-Color Fruits Using HSV （Hue, Saturation, Value） Color Space

Machine vision has been recently utilized for quality control of food and agricultural products, which was traditionally done by manual inspection. The present study was an attempt for automatic defect detection and sorting of some single-color fruits such as banana and plum. Fruit images were captured using a color digital camera with capturing direction of zero degree and under illuminant D65. It was observed that growing decay and time-aging made surface color changes in bruised parts of the object. 3D RGB and HSV color vectors as well as a single channel like H （hue）, S （saturation）, V （value） and grey scale images were applied for color quantization of the object. Results showed that there was a distinct threshold in the histogram of the S channel of images which can be applied to separate the object from its background. Moreover, the color change via the defect and time-aging is correctly distinguishable in the hue channel image. The effect of illumination, gloss and shadow of 3D image processing is less noticeable for hue data in comparison to saturation and value. The value of H channel was quantized to five groups based on the difference between each pixel value and the H value of a healthy object. The percentage of different degree of defects can be computed and used for grading the fruits.

Intelligent detection method for workpiece defect based on industrial CT image

In order to solve the problem of internal defect detection in industry, an intelligent detection method for workpiece defect based on industrial computed tomography (CT) images is proposed. The industrial CT slice image is preprocessed first with the combination of adaptive median filtering and adaptive weighted average filtering by analyzing the characteristics of the industrial CT slice images. Then an image segmentation algorithm based on gray change rate is used to segment low contrast information in industrial CT images, and the feature of workpiece defect is extracted by using Hu invariant moment. On this basis, the radial basis function (RBF) neural network model is established and the firefly algorithm is used for optimization, and the intelligent identification of the internal defects of the workpiece is completed. Simulation results show that this method can effectively improve the accuracy of defect identification and provide a theoretical basis for the detection of internal defects in industry.

Detection of surface cutting defect on magnet using Fourier image reconstruction

A magnet is an important component of a speaker,as it makes the coil move back forth,and it is commonly used in mobile information terminals.Defects may appear on the surface of the magnet while cutting it into smaller slices,and hence,automatic detection of surface cutting defect detection becomes an important task for magnet production.In this work,an image-based detection system for magnet surface defect was constructed,a Fourier image reconstruction based on the magnet surface image processing method was proposed.The Fourier transform was used to get the spectrum image of the magnet image,and the defect was shown as a bright line in it.The Hough transform was used to detect the angle of the bright line,and this line was removed to eliminate the defect from the original gray image;then the inverse Fourier transform was applied to get the background gray image.The defect region was obtained by evaluating the gray-level differences between the original image and the background gray image.Further,the effects of several parameters in this method were studied and the optimized values were obtained.Experiment results show that the proposed method can detect surface cutting defects in a magnet automatically and efficiently.