Research on image segmentation of Miao costume patterns based on the RSKP-UNet model北大核心CSCD

The Miao nationality is the sixth largest ethnic group in China which has a history of thousands of years. It has created a unique material culture and spiritual culture in its development process and the Miao costume is a highly condensed collection of the material and spiritual culture of the Miao nationality. As one of the unique symbols of Miao culture the Miao costume has profound cultural heritage and cultural connotations. The patterns of the Miao costume are particularly eye-catching as they not only symbolize the wisdom of the Miao people in thousands of years of production and life but also symbolize the pursuit of the good spirit of the Miao people. However under the impact of modern pop culture and foreign culture these cultural symbols are gradually disappearing. In order to protect and inherit them the Miao costume pattern segmentation has become the most important work. However the Miao costume pattern segmentation is quite difficult. At present there are few studies on the extraction classification identification and preservation of the features of Miao costume pattern segmentation. With the excellent segmentation performance of the U-Net model and the advantages of easy deployment the paper improves the basic structure of the U-Net model and proposes a Miao costume pattern segmentation algorithm based on the RSKP-UNet Residual Selective-Kernel Parallel U-Net model. The algorithm adds Residual modules in the encoder part of the U-Net model to improve the feature extraction capability of the model and embeds the SKNet modules and ParNet modules in the decoder part to enhance the feature expression capability of the model. The paper uses evaluation metrics to measure the segmentation performance of the model and compares it with four segmentation models based on deep learning. The paper not only combines the current research focus-deep learning and attention mechanism but also introduces the Lovász-hinge loss function to effectively solve the problem of class imbalance in the Miao costume patterns. The RSKP-UNet model is better than other models in various segmentation indicators. Compared with the benchmark model U-Net the Dice coefficient IoU precision recall and accuracy are improved by 6. 98% 11. 07% 2. 89% 6. 75% and 3. 92% . The segmentation algorithm proposed in this paper realizes the extraction of the element content of the Miao costume patterns through image segmentation of Miao costume patterns which can be used to build the Miao costume pattern database in this way thus helping designers relevant researchers and organizations to provide research foundation and completing the protection and inheritance of the Miao costume culture. The paper also provides some reference for the segmentation research of other minority costume patterns. © 2022 Authors. All rights reserved.     

GCAT-U-Net嵌入全局坐标注意力机制的遥感地块分割网络

遥感影像的地块背景特征复杂,当前地块分割方法不能较好地处理模糊的边缘信息,导致分割精度不理想;文章利用注意力机制处理地块特征,提出了一种基于全局坐标注意力机制的遥感地块分割网络:GCAT-U-Net;该方法在U-Net网络基础上嵌入了全局坐标注意力机制,加强了深度神经网络对于遥感影像数据中重要特征的关注度;在公开的GID数据集上的实验结果表明,文章提出的模型将准确率从0.9041提升到了0.9227,比传统U-Net网络提高了2百分点;结合特征自身重要性和特征位置信息的全局坐标注意力机制有助于更精确的目标定位,其输出相较于嵌入单一注意力机制,地块边界更为清晰,提升效果更为显著。     

检测小篡改区域的U型网络

目的 图像篡改区域检测是图像取证领域的一个挑战性任务,其目的是找出图像的篡改区域。传统方法仅针对某种特定的篡改方式进行设计,难以检测其他篡改方式的图像。基于卷积神经网络的方法能够自适应地提取特征,同时检测包含多种篡改方式的图像。但是其中多数方法都选择增强图像的噪声特征,这种机制无法较好处理篡改区域与原图像来源相同、噪声相似的情况。多数方法还忽略了篡改区域过小而产生的样本不平衡问题,导致检测效果不佳。方法 提出了一个基于区域损失的用于检测小篡改区域的U型网络,该网络构建了一个异常区域特征增强机制,放大与图像背景差异较大的异常区域的特征。此外,还利用区域损失增强对篡改区域框内像素的判别能力,可以解决因篡改区域过小而产生的样本不平衡问题。结果 消融实验说明了异常区域特征增强机制和区域损失机制的有效性;对JPEG压缩和高斯模糊的对抗性测试证明了模型的鲁棒性;在CASIA2.0（CASI-A image tampering detection evaluation database）、NIST2016（NIST nimble 2016 datasets）、COLUMBIA （Columbia uncompressed image splicing detection evaluation dataset）和COVERAGE （a novel database forcopy-move forgery detection）数据集上与最新方法进行比较时,本文方法取得了最优性能,其F1 score分别为0.979 5、0.982 2、0.995 3和0.987 0。结论 本文的异常区域特征增强机制和区域损失机制能有效提高模型性能,同时缓解篡改区域过小导致的样本不平衡问题,大量实验也表明了本文提出的小篡改区域检测方法的优越性。     

基于U-Net的列车轮对激光曲线提取

研究列车轮对条纹图像快速准确提取的方法,采用经典的U-Net网络模型,实现了激光条纹的精确分割,以构建模板的方式对分割后的图像采用灰度重心法达到亚像素的提取。首先利用U-Net网络模型对激光条纹进行分割,然后用模板法初步找到光条中心,最后再使用灰度重心法实现快速、准确的激光曲线提取。结果表明,该方法可以有效地克服动态环境下背景噪声以及亮斑对激光条纹提取带来的影响。     

基于深度学习的铸件 CT 图像分割算法

针对现有方法分割弱边缘铸件 CT 图像难度大、精度低、鲁棒性差的问题,提出一种融合残差模块与混合注意力机制的 U 型网络分割算法(AttRes-U-Nets)。 该算法以 U-Net 网络为基础,首先构建深度残差网络 ResNets 作为算法的编码网络,解决 传统 U-Net 网络特征提取能力不足的问题;然后,引入改进后的混合注意力机制,突出分割目标区域与通道的特征响应,提高网 络灵敏度;最后,将 Focal loss 与 Dice loss 结合为一种新损失函数 FD loss 缓解样本不平衡带来的负面影响。 使用 120 阀体数据 集对算法性能进行验证,实验结果表明,本文算法对铸件分割的像素准确率(PA)和交互比( IoU)分别达到 98. 72% 和 97. 40% , 优于传统 U-Net 算法与其他主流语义分割算法,为弱边缘分割提供了新思路。     

基于物理约束U-Net网络的地震数据低频延拓

由于震源与采集技术的影响,地震勘探数据往往缺失低频信息,严重影响后续的反演和成像处理。现有的地震数据低频延拓方法,大多建立在时域数据分布特征的基础上,容易导致频率与相位信息严重损失。为解决该问题,文中提出一种基于地震波物理参数约束的U-Net深度学习网络进行地震数据低频延拓。首先,利用理论引导数据的思想组织样本,生成大量不同特征的地震数据;然后,通过结合残差跳跃连接改进的U-Net模型学习从中高频地震数据生成低频成分的非线性映射;最后,结合地震信号的物理参数约束提升对频率、相位的恢复效果。实验证明,文中所提方法对地震数据低频恢复具有较好的效果,并在频率与相位的保持上优于同类方法,对提高后续的处理与解释精度具有较高的实用价值。     

基于图像分析和深度学习的复合绝缘子憎水性分级

为了更加方便快捷地检测大量复合绝缘子憎水性等级,提出一种基于图像分析和深度学习的复合绝缘子憎水性分级方法。首先为提高图像对比度,对复合绝缘子憎水性图像进行灰度化和图像增强处理;其次利用图像分析技术和U-Net网络提取水珠轮廓,得到水珠轮廓图像;接着引入深度卷积神经网络,将这些水珠轮廓图作为神经网络的输入,以相应的憎水性等级作为输出向量,训练网络得到分级模型;最后将分级模型用于憎水性分级,得到分级结果。实验结果表明:该方法的分级结果已达到实际应用要求,水珠轮廓提取的精度达到了92.96%,分级准确率达到了90.2%,预测一幅图像的憎水性等级平均耗时0.1 s。     

Empirical Comparisons of Deep Learning Networks on Liver Segmentation

Accurate segmentation of CT images of liver tumors is an important adjunct for the liver diagnosis and treatment of liver diseases. In recent years, due to the great improvement of hard device, many deep learning based methods have been proposed for automatic liver segmentation. Among them, there are the plain neural network headed by FCN and the residual neural network headed by Resnet, both of which have many variations. They have achieved certain achievements in medical image segmentation. In this paper, we firstly select five representative structures, i.e., FCN, U-Net, Segnet, Resnet and Densenet, to investigate their performance on liver segmentation. Since original Resnet and Densenet could not perform image segmentation directly, we make some adjustments for them to perform live segmentation. Our experimental results show that Densenet performs the best on liver segmentation, followed by Resnet. Both perform much better than Segnet, U-Net, and FCN. Among Segnet, U-Net, and FCN, U-Net performs the best, followed by Segnet. FCN performs the worst.     

基于改进Mask TextSpotterV3模型的钢卷标号识别

本文以弯曲超长的钢卷标号为研究对象,提出一种调整字符区域方位的预处理方法和钢卷标号识别的方案。首先使用改进U-Net 模型分割字符区域并将文本区域调整至图像上方,然后使用改进Mask TextSpotter V3进行标号识别。实验结果表明,相较于原模型,改进的U-Net模型在反光等干扰的情况下提升了字符区域分割效果,交并比提升了14%;改进的Mask TextSpotter V3模型改善了原模型易受噪声影响、模糊图像文本检测准确率低的问题,在整体长字符检测与识别中,模糊图像的文本检测召回率提高了4%,字符错误率降低了12.49%。整体模型框架针对变长弯曲的多方向长字符在真实场景中准确率能够稳定在98%及以上,在对钢卷标号或类似对象的识别上有一定的先进性、实用性和参考性。     

基于空洞U-Net网络的乳腺细胞图像分割算法

乳腺细胞的准确分割是乳腺组织切片图像病理分 析的关键环节,对乳腺癌的诊治具有 重要价值。针对乳腺细胞图像分割中细胞边界不清晰、分割准确率低的问题,提出一种基于 空洞U-Net网络的乳腺细胞图像分割算法。在U-Net网络中引入空洞卷积增大卷积层感受野 , 获得包含更多乳腺细胞边缘信息的特征图,在卷积层和池化层间增加实例归一化层,提高网 络收敛速度的同时缓解过拟合现象,并使用加权损失函数增强乳腺细胞区域的权重,提高网 络对细胞特征的提取能力,实现乳腺细胞边界的有效分割。实验结果表明,该算法在USCB B reast数据集上的分割准确率和Dice系数分别达到97.63%和83.25%,较原始U-Net网络分别提高了6.5%,对乳腺细胞图像具有更好的分割效果。