非结构化道路的缺陷目标检测任务对道路交通安全具有重要意义,但检测所需的标注数据集相对有限。为了解决非结构化道路标注数据集缺乏以及现有模型对无标注数据学习能力不足的问题,提出一种MAM(Multi-Augmentation with Memory)半监督...非结构化道路的缺陷目标检测任务对道路交通安全具有重要意义,但检测所需的标注数据集相对有限。为了解决非结构化道路标注数据集缺乏以及现有模型对无标注数据学习能力不足的问题,提出一种MAM(Multi-Augmentation with Memory)半监督目标检测算法。首先,引入缓存机制存储无标注图像和带有伪标注图像的框回归位置信息,避免了后续匹配造成的计算资源浪费。其次,设计混合数据增强策略,将缓存的伪标签图像与无标签图像混合输入学生模型,以增强模型对新数据的泛化能力,并使图像的尺度分布更加均衡。MAM算法不受目标检测模型的限制,并且更好地保持了目标框的一致性,避免了计算一致性损失。实验结果表明,MAM算法相比其他全监督学习和半监督学习算法更具优越性,在自建的非结构化道路缺陷数据集Defect上,在标注比例为10%、20%和30%的场景下,MAM算法的均值平均精度(mAP)相比于Soft Teacher算法分别提升了6.8、11.1和6.0百分点,在自建的非结构化道路坑洼数据集Pothole上,在标注比例为15%和30%的场景下,MAM算法的mAP相比于Soft Teacher算法分别提升了5.8和4.3百分点。展开更多
Baosteel developed a digital automatic analysis technique for maceral specification in 2002. This analysis system combines digital image processing, graphics, databases, expert systems, artificial intelligence and oth...Baosteel developed a digital automatic analysis technique for maceral specification in 2002. This analysis system combines digital image processing, graphics, databases, expert systems, artificial intelligence and other advanced technologies. After 6 years of application in coke production, the system proved itself successful in coal quality testing and coal blending guidance on maceral. However,during this long process, some inadequacies were found that impacted the precision and accuracy of the analysis. So ,in 2008 Baosteel began to work on improving the coal maceral analysis system. The improvements included the following：further upgrading and enhancing the analysis performance of microscopic images ;extending the gray levels to increase the reflectance measurement accuracy 64 times;changing the focus method and effectively eliminating the interference of halo. In addition, an improved image recognition method was adopted to make the extraction of vitrinite more accurate and a new model of coal constituent algorithm was added which can accurately determine the composition of maceral （exinite, vitrinite,inertinite）. Since these improvements were completed, the system has achieved higher automation, speed and accuracy, collected more information and performed more accurate maceral analysis for coke production. Meanwhile, the improved system has provided a reliable analytical basis for the further study on the relationship between coke quality and coal blending.展开更多
文摘非结构化道路的缺陷目标检测任务对道路交通安全具有重要意义,但检测所需的标注数据集相对有限。为了解决非结构化道路标注数据集缺乏以及现有模型对无标注数据学习能力不足的问题,提出一种MAM(Multi-Augmentation with Memory)半监督目标检测算法。首先,引入缓存机制存储无标注图像和带有伪标注图像的框回归位置信息,避免了后续匹配造成的计算资源浪费。其次,设计混合数据增强策略,将缓存的伪标签图像与无标签图像混合输入学生模型,以增强模型对新数据的泛化能力,并使图像的尺度分布更加均衡。MAM算法不受目标检测模型的限制,并且更好地保持了目标框的一致性,避免了计算一致性损失。实验结果表明,MAM算法相比其他全监督学习和半监督学习算法更具优越性,在自建的非结构化道路缺陷数据集Defect上,在标注比例为10%、20%和30%的场景下,MAM算法的均值平均精度(mAP)相比于Soft Teacher算法分别提升了6.8、11.1和6.0百分点,在自建的非结构化道路坑洼数据集Pothole上,在标注比例为15%和30%的场景下,MAM算法的mAP相比于Soft Teacher算法分别提升了5.8和4.3百分点。
文摘Baosteel developed a digital automatic analysis technique for maceral specification in 2002. This analysis system combines digital image processing, graphics, databases, expert systems, artificial intelligence and other advanced technologies. After 6 years of application in coke production, the system proved itself successful in coal quality testing and coal blending guidance on maceral. However,during this long process, some inadequacies were found that impacted the precision and accuracy of the analysis. So ,in 2008 Baosteel began to work on improving the coal maceral analysis system. The improvements included the following：further upgrading and enhancing the analysis performance of microscopic images ;extending the gray levels to increase the reflectance measurement accuracy 64 times;changing the focus method and effectively eliminating the interference of halo. In addition, an improved image recognition method was adopted to make the extraction of vitrinite more accurate and a new model of coal constituent algorithm was added which can accurately determine the composition of maceral （exinite, vitrinite,inertinite）. Since these improvements were completed, the system has achieved higher automation, speed and accuracy, collected more information and performed more accurate maceral analysis for coke production. Meanwhile, the improved system has provided a reliable analytical basis for the further study on the relationship between coke quality and coal blending.
