针对印刷机轴承套筒装配过程中无约束运动阶段位姿测量问题,以Stewart平台为装配操作平台,通过提取双目相机下的图像椭圆特征,设计出一种基于机器视觉的印刷机轴套工装位姿测量系统。首先以轴套工装外径与相机成像平面之间投影形成的椭...针对印刷机轴承套筒装配过程中无约束运动阶段位姿测量问题,以Stewart平台为装配操作平台,通过提取双目相机下的图像椭圆特征,设计出一种基于机器视觉的印刷机轴套工装位姿测量系统。首先以轴套工装外径与相机成像平面之间投影形成的椭圆特征为识别目标,利用投影几何关系估算轴套工装位姿得到二义性解,用双目视觉定位技术剔除伪解,最后根据位姿变换得到工装位姿并且反解出六组支杆杆长。实验结果表明,杆长最大误差率不超过1.84%,能为无约束运动阶段轴孔装配提供位姿信息和杆长控制,进而实现印刷机轴套装配中无约束运动阶段的轴孔装配。Aiming at the pose measurement problem during the unconstrained motion stage of the printing machine bearing sleeve assembly process, using the Stewart platform as the assembly operation platform, a machine vision based printing machine shaft sleeve tooling pose measurement system is designed by extracting image elliptical features under binocular cameras. First and for most, estimate the pose of this sleeve fixture using geometric relationships. There will be two pose solutions in this step. Then, use binocular vision positioning technology to eliminate erroneous pose solutions. Finally, based on the pose transformation, the fixture pose is obtained and six sets of support rod lengths are solved in reverse. The experimental results show that the maximum error rate of the rod length does not exceed 1.84%, which can provide pose information and rod length control for the assembly of the shaft hole in the unconstrained motion stage. Thus, it achieves the assembly of the shaft hole in the unconstrained motion stage of the printing machine shaft assembly.展开更多
传统的图像隐写往往倾向于将隐藏信息安全地嵌入到封面图像中,而几乎忽略了有效负载容量。为解决传统隐写容量低的问题,本文采用深度学习与图像信息隐藏相结合的方法。实验结果表明,在嵌入容量上,所提算法达到了24 bpp,是目前容量最大...传统的图像隐写往往倾向于将隐藏信息安全地嵌入到封面图像中,而几乎忽略了有效负载容量。为解决传统隐写容量低的问题,本文采用深度学习与图像信息隐藏相结合的方法。实验结果表明,在嵌入容量上,所提算法达到了24 bpp,是目前容量最大的图像隐写算法之一。在此大容量嵌入的前提下,所提算法生成的载密图像和提取的秘密图像,无论在主观视觉质量还是客观视觉指标峰值信噪比(PSNR)上都高于其他同类算法,说明了设计的端到端隐写网络的整体优越性。Traditional image steganography methods often focus on securely embedding hidden information into cover images, while paying little attention to the payload capacity. To address the issue of low embedding capacity in conventional steganography, this paper combines deep learning with image information hiding techniques. Experimental results show that the proposed algorithm achieves an embedding capacity of 24 bpp, making it one of the highest-capacity image steganography algorithms to date. Despite the large embedding capacity, the stego-images generated by the algorithm and the extracted secret images outperform other similar algorithms in both subjective visual quality and objective visual metrics such as Peak Signal-to-Noise Ratio (PSNR). This demonstrates the overall superiority of the designed end-to-end steganography network.展开更多
为了解决Shamir(t, n)门限方案在秘密共享时,未能充分利用多项式系数和共享份额的问题,本文设计了一种独立高容量半色调图像信息隐藏算法。利用多项式的常数项和一次项系数隐藏秘密图像,共享份额隐藏份额编号或者用户信息等单个共享份...为了解决Shamir(t, n)门限方案在秘密共享时,未能充分利用多项式系数和共享份额的问题,本文设计了一种独立高容量半色调图像信息隐藏算法。利用多项式的常数项和一次项系数隐藏秘密图像,共享份额隐藏份额编号或者用户信息等单个共享份额信息。利用二次项系数隐藏版权信息或者防伪信息。单个共享份额和达到门限数量的多个共享份额都可以进行认证。实验表明,该算法可分离秘密图像和多个共享份额的信息,实现多种信息的可逆隐藏,并且利用单个份额隐藏单个份额的特有信息。该算法增加了信息的种类和嵌入容量。对于Shamir(3, 5)门限共享,嵌入率可以达到3.5 bpp。In order to solve the problem of Shamir(t, n) threshold scheme not fully utilizing polynomial coefficients and shared shares during secret sharing, an independent high-capacity halftone image information hiding algorithm was designed in this paper. Using the constant term and first-order coefficient of polynomials to hide secret images, sharing shares to hide individual shared share information, such as share numbers or user information. Using quadratic coefficients to hide copyright or anti-counterfeiting information. Single shared shares and multiple shared shares that reach the threshold can be authenticated. Experiments have shown that this algorithm can separate secret images and information from multiple shared shares, achieve reversible hiding of multiple types of information, and utilize a single share to hide unique information of a single share. This algorithm increases the variety and embedding capacity of information. For Shamir(3, 5) threshold sharing, the embedding rate can reach 3.5 bpp.展开更多
为了解决当前国内市场中多色套印技术在曲面丝网印刷中的精确对准难题,研究提出了一种基于ORB图像匹配算法的多色套印系统,该系统结合了几何约束和RANSAC筛选算法,使传统曲面丝网印刷不再局限于单色制品,满足了市场对高质量、多色印刷...为了解决当前国内市场中多色套印技术在曲面丝网印刷中的精确对准难题,研究提出了一种基于ORB图像匹配算法的多色套印系统,该系统结合了几何约束和RANSAC筛选算法,使传统曲面丝网印刷不再局限于单色制品,满足了市场对高质量、多色印刷日益增长的需求。通过实验验证了系统的有效性,结果表明匹配正确率超过95%,处理速度达26帧/秒,色板套印精度控制在± 0.03毫米以内。这种方法显著提升了复杂多色图案在曲面物体上的印刷质量和精度,具备广泛的应用前景。To address the challenges of precise alignment in multi-color overprinting for curved surface screen printing in the domestic market, this study proposes a multi-color overprinting system based on the ORB image matching algorithm, incorporating geometric constraints and the RANSAC filtering algorithm. This approach extends traditional curved surface screen printing beyond single-color applications, meeting the increasing market demand for high-quality, multi-color printing. Experimental results verify the system’s effectiveness, demonstrating a matching accuracy rate exceeding 95%, a processing speed of 26 frames per second, and an overprinting accuracy within ± 0.03 mm. This method significantly enhances the print quality and precision of complex multi-color patterns on curved objects, with broad application potential.展开更多
随着信息技术的快速发展,数据安全问题日益受到重视。本文提出了一种可验证的秘密分享方案,该方案基于Shamir秘密分享方案,并结合NTRU数字签名算法,增强方案的安全性。NTRU数字签名算法作为一种能够抵抗量子攻击的数字签名算法,有效防...随着信息技术的快速发展,数据安全问题日益受到重视。本文提出了一种可验证的秘密分享方案,该方案基于Shamir秘密分享方案,并结合NTRU数字签名算法,增强方案的安全性。NTRU数字签名算法作为一种能够抵抗量子攻击的数字签名算法,有效防御了伪造和篡改攻击,确保了秘密恢复过程的可信度。本文详细分析了方案的正确性和安全性。With the rapid development of information technology, data security issues are increasingly being taken seriously. This paper proposes a verifiable secret sharing scheme based on the Shamir secret sharing scheme and combined with the NTRU digital signature algorithm to enhance the security of the scheme. The NTRU digital signature algorithm, as a type of digital signature algorithm capable of resisting quantum attacks, effectively defends against forgery and tampering attacks, ensuring the credibility of the secret recovery process. This paper provides a detailed analysis of the correctness and security of the scheme.展开更多
随着人工智能技术的迅猛发展,医疗问答系统已成为医疗信息检索和知识获取的重要工具。医疗领域涉及大量医学术语、复杂的疾病症状和治疗方案,传统查询方式难以高效、准确地满足医护人员和患者的信息需求。相比传统国内搜索引擎和原生开...随着人工智能技术的迅猛发展,医疗问答系统已成为医疗信息检索和知识获取的重要工具。医疗领域涉及大量医学术语、复杂的疾病症状和治疗方案,传统查询方式难以高效、准确地满足医护人员和患者的信息需求。相比传统国内搜索引擎和原生开源大语言模型(LLMs),基于LangChain的大模型医疗问答系统能够提供更高质量的答案,显著提升医疗知识检索的效率和精准度。因此,本研究提出了一种基于LangChain与大模型的医疗智能问答系统,结合命名实体识别(NER)、图谱查询和对话分析等技术,构建了一个专注于医疗领域的知识图谱及其查询与生成模块。通过设计和优化Prompt提示词,Agent Tool提升了大模型生成更精准、高质量医疗问答的能力。研究结果表明,该系统在医疗问答任务中的表现优异,准确度、方案可行性和上下文相关性等指标显著优于传统LLMs和国内知名大模型。该系统通过与大规模医疗知识图谱的结合,能够深入理解复杂的医疗问题,并提供精准的回答,呈现可视化图谱展示图,更直观地给用户反馈,同时具备较高的数据安全性和可迁移性。Nowadays, with the rapid development of artificial intelligence technology, medical question answering system has become an important tool for medical information retrieval and knowledge acquisition. The medical field involves a large number of medical terms, complicated disease symptoms and treatment plans, and traditional inquiry methods are difficult to meet the information needs of medical staff and patients efficiently and accurately. Compared with traditional domestic search engines and native open source large language model (LLMs), LangChain-based large model medical question answering system can provide higher quality answers, significantly improving the efficiency and accuracy of medical knowledge retrieval. Therefore, this study proposed a medical intelligent question and answer system based on LangChain and large model, combined with named entity recognition (NER), graph query and dialogue analysis and other technologies, to build a knowledge graph and query and generation module focusing on the medical field. By designing and optimizing Prompt words, Agent Tool improves the ability of large models to generate more accurate and high-quality medical questions and answers. The results show that the system performs well in medical question answering tasks, with significant improvements in accuracy, feasibility, and context relevance are significantly better than traditional LLMs and well-known domestic large models. Through the combination of large-scale medical knowledge graph, the system can deeply understand complex medical questions, provide accurate answers, present a visual map display graph, and give users more intuitive feedback, while having high data security and portability.展开更多
恶劣天气环境下拍摄的图像会受到雾或霾的影响,从而导致图像饱和度过低模糊、以及颜色灰白等负面效果,这不仅会使图像中的重要信息丢失,还会对后续计算机视觉任务(如目标检测、图像分割、人员再识别)的研究造成负面影响。为了解决上述问...恶劣天气环境下拍摄的图像会受到雾或霾的影响,从而导致图像饱和度过低模糊、以及颜色灰白等负面效果,这不仅会使图像中的重要信息丢失,还会对后续计算机视觉任务(如目标检测、图像分割、人员再识别)的研究造成负面影响。为了解决上述问题,文章首先对图像去雾的发展历程进行分析和梳理,接下来重点论述了深度学习在图像去雾领域的研究进展,主要包含有监督去雾、无监督去雾和半监督去雾技术,并对各自的代表性算法进行深入对比分析。最后,介绍了图像去雾领域主流的数据集和评估指标。Images captured in harsh weather environments are often affected by fog or haze, which can lead to negative effects such as low saturation, blurring, and grayish-white colors. This not only results in the loss of important information in the image, but also has a negative impact on subsequent computer vision tasks such as object detection, image segmentation, and personnel re-identification. This article first provides a comprehensive analysis and sorting of image defogging and then reviews the research progress of deep learning in the field of image defogging, mainly including supervised defogging, unsupervised defogging, and semi-supervised defogging. We compared and analyzed representative algorithms among these methods. Finally, the commonly used datasets and evaluation metrics for image defogging were introduced.展开更多
文摘针对印刷机轴承套筒装配过程中无约束运动阶段位姿测量问题,以Stewart平台为装配操作平台,通过提取双目相机下的图像椭圆特征,设计出一种基于机器视觉的印刷机轴套工装位姿测量系统。首先以轴套工装外径与相机成像平面之间投影形成的椭圆特征为识别目标,利用投影几何关系估算轴套工装位姿得到二义性解,用双目视觉定位技术剔除伪解,最后根据位姿变换得到工装位姿并且反解出六组支杆杆长。实验结果表明,杆长最大误差率不超过1.84%,能为无约束运动阶段轴孔装配提供位姿信息和杆长控制,进而实现印刷机轴套装配中无约束运动阶段的轴孔装配。Aiming at the pose measurement problem during the unconstrained motion stage of the printing machine bearing sleeve assembly process, using the Stewart platform as the assembly operation platform, a machine vision based printing machine shaft sleeve tooling pose measurement system is designed by extracting image elliptical features under binocular cameras. First and for most, estimate the pose of this sleeve fixture using geometric relationships. There will be two pose solutions in this step. Then, use binocular vision positioning technology to eliminate erroneous pose solutions. Finally, based on the pose transformation, the fixture pose is obtained and six sets of support rod lengths are solved in reverse. The experimental results show that the maximum error rate of the rod length does not exceed 1.84%, which can provide pose information and rod length control for the assembly of the shaft hole in the unconstrained motion stage. Thus, it achieves the assembly of the shaft hole in the unconstrained motion stage of the printing machine shaft assembly.
文摘传统的图像隐写往往倾向于将隐藏信息安全地嵌入到封面图像中,而几乎忽略了有效负载容量。为解决传统隐写容量低的问题,本文采用深度学习与图像信息隐藏相结合的方法。实验结果表明,在嵌入容量上,所提算法达到了24 bpp,是目前容量最大的图像隐写算法之一。在此大容量嵌入的前提下,所提算法生成的载密图像和提取的秘密图像,无论在主观视觉质量还是客观视觉指标峰值信噪比(PSNR)上都高于其他同类算法,说明了设计的端到端隐写网络的整体优越性。Traditional image steganography methods often focus on securely embedding hidden information into cover images, while paying little attention to the payload capacity. To address the issue of low embedding capacity in conventional steganography, this paper combines deep learning with image information hiding techniques. Experimental results show that the proposed algorithm achieves an embedding capacity of 24 bpp, making it one of the highest-capacity image steganography algorithms to date. Despite the large embedding capacity, the stego-images generated by the algorithm and the extracted secret images outperform other similar algorithms in both subjective visual quality and objective visual metrics such as Peak Signal-to-Noise Ratio (PSNR). This demonstrates the overall superiority of the designed end-to-end steganography network.
文摘为了解决Shamir(t, n)门限方案在秘密共享时,未能充分利用多项式系数和共享份额的问题,本文设计了一种独立高容量半色调图像信息隐藏算法。利用多项式的常数项和一次项系数隐藏秘密图像,共享份额隐藏份额编号或者用户信息等单个共享份额信息。利用二次项系数隐藏版权信息或者防伪信息。单个共享份额和达到门限数量的多个共享份额都可以进行认证。实验表明,该算法可分离秘密图像和多个共享份额的信息,实现多种信息的可逆隐藏,并且利用单个份额隐藏单个份额的特有信息。该算法增加了信息的种类和嵌入容量。对于Shamir(3, 5)门限共享,嵌入率可以达到3.5 bpp。In order to solve the problem of Shamir(t, n) threshold scheme not fully utilizing polynomial coefficients and shared shares during secret sharing, an independent high-capacity halftone image information hiding algorithm was designed in this paper. Using the constant term and first-order coefficient of polynomials to hide secret images, sharing shares to hide individual shared share information, such as share numbers or user information. Using quadratic coefficients to hide copyright or anti-counterfeiting information. Single shared shares and multiple shared shares that reach the threshold can be authenticated. Experiments have shown that this algorithm can separate secret images and information from multiple shared shares, achieve reversible hiding of multiple types of information, and utilize a single share to hide unique information of a single share. This algorithm increases the variety and embedding capacity of information. For Shamir(3, 5) threshold sharing, the embedding rate can reach 3.5 bpp.
文摘为了解决当前国内市场中多色套印技术在曲面丝网印刷中的精确对准难题,研究提出了一种基于ORB图像匹配算法的多色套印系统,该系统结合了几何约束和RANSAC筛选算法,使传统曲面丝网印刷不再局限于单色制品,满足了市场对高质量、多色印刷日益增长的需求。通过实验验证了系统的有效性,结果表明匹配正确率超过95%,处理速度达26帧/秒,色板套印精度控制在± 0.03毫米以内。这种方法显著提升了复杂多色图案在曲面物体上的印刷质量和精度,具备广泛的应用前景。To address the challenges of precise alignment in multi-color overprinting for curved surface screen printing in the domestic market, this study proposes a multi-color overprinting system based on the ORB image matching algorithm, incorporating geometric constraints and the RANSAC filtering algorithm. This approach extends traditional curved surface screen printing beyond single-color applications, meeting the increasing market demand for high-quality, multi-color printing. Experimental results verify the system’s effectiveness, demonstrating a matching accuracy rate exceeding 95%, a processing speed of 26 frames per second, and an overprinting accuracy within ± 0.03 mm. This method significantly enhances the print quality and precision of complex multi-color patterns on curved objects, with broad application potential.
文摘随着信息技术的快速发展,数据安全问题日益受到重视。本文提出了一种可验证的秘密分享方案,该方案基于Shamir秘密分享方案,并结合NTRU数字签名算法,增强方案的安全性。NTRU数字签名算法作为一种能够抵抗量子攻击的数字签名算法,有效防御了伪造和篡改攻击,确保了秘密恢复过程的可信度。本文详细分析了方案的正确性和安全性。With the rapid development of information technology, data security issues are increasingly being taken seriously. This paper proposes a verifiable secret sharing scheme based on the Shamir secret sharing scheme and combined with the NTRU digital signature algorithm to enhance the security of the scheme. The NTRU digital signature algorithm, as a type of digital signature algorithm capable of resisting quantum attacks, effectively defends against forgery and tampering attacks, ensuring the credibility of the secret recovery process. This paper provides a detailed analysis of the correctness and security of the scheme.
文摘随着人工智能技术的迅猛发展,医疗问答系统已成为医疗信息检索和知识获取的重要工具。医疗领域涉及大量医学术语、复杂的疾病症状和治疗方案,传统查询方式难以高效、准确地满足医护人员和患者的信息需求。相比传统国内搜索引擎和原生开源大语言模型(LLMs),基于LangChain的大模型医疗问答系统能够提供更高质量的答案,显著提升医疗知识检索的效率和精准度。因此,本研究提出了一种基于LangChain与大模型的医疗智能问答系统,结合命名实体识别(NER)、图谱查询和对话分析等技术,构建了一个专注于医疗领域的知识图谱及其查询与生成模块。通过设计和优化Prompt提示词,Agent Tool提升了大模型生成更精准、高质量医疗问答的能力。研究结果表明,该系统在医疗问答任务中的表现优异,准确度、方案可行性和上下文相关性等指标显著优于传统LLMs和国内知名大模型。该系统通过与大规模医疗知识图谱的结合,能够深入理解复杂的医疗问题,并提供精准的回答,呈现可视化图谱展示图,更直观地给用户反馈,同时具备较高的数据安全性和可迁移性。Nowadays, with the rapid development of artificial intelligence technology, medical question answering system has become an important tool for medical information retrieval and knowledge acquisition. The medical field involves a large number of medical terms, complicated disease symptoms and treatment plans, and traditional inquiry methods are difficult to meet the information needs of medical staff and patients efficiently and accurately. Compared with traditional domestic search engines and native open source large language model (LLMs), LangChain-based large model medical question answering system can provide higher quality answers, significantly improving the efficiency and accuracy of medical knowledge retrieval. Therefore, this study proposed a medical intelligent question and answer system based on LangChain and large model, combined with named entity recognition (NER), graph query and dialogue analysis and other technologies, to build a knowledge graph and query and generation module focusing on the medical field. By designing and optimizing Prompt words, Agent Tool improves the ability of large models to generate more accurate and high-quality medical questions and answers. The results show that the system performs well in medical question answering tasks, with significant improvements in accuracy, feasibility, and context relevance are significantly better than traditional LLMs and well-known domestic large models. Through the combination of large-scale medical knowledge graph, the system can deeply understand complex medical questions, provide accurate answers, present a visual map display graph, and give users more intuitive feedback, while having high data security and portability.
文摘恶劣天气环境下拍摄的图像会受到雾或霾的影响,从而导致图像饱和度过低模糊、以及颜色灰白等负面效果,这不仅会使图像中的重要信息丢失,还会对后续计算机视觉任务(如目标检测、图像分割、人员再识别)的研究造成负面影响。为了解决上述问题,文章首先对图像去雾的发展历程进行分析和梳理,接下来重点论述了深度学习在图像去雾领域的研究进展,主要包含有监督去雾、无监督去雾和半监督去雾技术,并对各自的代表性算法进行深入对比分析。最后,介绍了图像去雾领域主流的数据集和评估指标。Images captured in harsh weather environments are often affected by fog or haze, which can lead to negative effects such as low saturation, blurring, and grayish-white colors. This not only results in the loss of important information in the image, but also has a negative impact on subsequent computer vision tasks such as object detection, image segmentation, and personnel re-identification. This article first provides a comprehensive analysis and sorting of image defogging and then reviews the research progress of deep learning in the field of image defogging, mainly including supervised defogging, unsupervised defogging, and semi-supervised defogging. We compared and analyzed representative algorithms among these methods. Finally, the commonly used datasets and evaluation metrics for image defogging were introduced.
