The traditional oriented FAST and rotated BRIEF（ORB） algorithm has problems of instability and repetition of keypoints and it does not possess scale invariance. In order to deal with these drawbacks, a modified ORB...The traditional oriented FAST and rotated BRIEF（ORB） algorithm has problems of instability and repetition of keypoints and it does not possess scale invariance. In order to deal with these drawbacks, a modified ORB（MORB） algorithm is proposed. In order to improve the precision of matching and tracking, this paper puts forward an MOK algorithm that fuses MORB and Kanade-Lucas-Tomasi（KLT）. By using Kalman, the object＇s state in the next frame is predicted in order to reduce the size of search window and improve the real-time performance of object tracking. The experimental results show that the MOK algorithm can accurately track objects with deformation or with background clutters, exhibiting higher robustness and accuracy on diverse datasets. Also, the MOK algorithm has a good real-time performance with the average frame rate reaching 90.8 fps.展开更多
为提高自然环境中玉米害虫识别的准确性,开发一种基于优化YOLOv8的深度学习模型YOLOv8-LAP。该模型将大型可分离卷积核注意力(LSKA)机制引入特征融合模块空间快速金字塔池化(SPPF),增强多尺度特征提取能力,提升检测性能。针对玉米害虫...为提高自然环境中玉米害虫识别的准确性,开发一种基于优化YOLOv8的深度学习模型YOLOv8-LAP。该模型将大型可分离卷积核注意力(LSKA)机制引入特征融合模块空间快速金字塔池化(SPPF),增强多尺度特征提取能力,提升检测性能。针对玉米害虫图像检测中小目标难以捕捉、背景复杂和光照变化等挑战,在主干网络中加入AFGC(Attention for Fine-Grained Categorization)层,以进一步增强图像特征提取的效果,提升模型的泛化能力和鲁棒性。为保证实时检测和模型轻量化,引入可编程梯度信息(PGI)技术,通过辅助监督优化训练过程,减少参数并加速推理。在9种常见玉米害虫的检测中,YOLOv8-LAP模型的平均精度均值(mAP0.5)达到了95.7%,相较于原始YOLOv8模型提高了4.9个百分点。此外,为验证YOLOv8-LAP模型的效果,开发一款基于PySide6的应用程序,该应用拥有用户友好的图形用户界面(GUI),具有实时图像处理和视频分析功能,并支持静态图像、动态视频和摄像头实时目标检测。可见,YOLOv8-LAP模型在降低漏检率和误检率方面表现突出,目标定位更精准,适用于自然环境下的玉米害虫识别,并为精准施药提供技术支持。展开更多
基于深度学习进行信号自动调制识别在分类精度、可迁移性等方面普遍优于传统方法,引起广泛关注。但是,当前方法多数针对单信号样本进行识别,无法适用于混叠信号识别场景。针对该问题,对混叠信号调制识别方法进行了研究,结合长短期记忆(l...基于深度学习进行信号自动调制识别在分类精度、可迁移性等方面普遍优于传统方法,引起广泛关注。但是,当前方法多数针对单信号样本进行识别,无法适用于混叠信号识别场景。针对该问题,对混叠信号调制识别方法进行了研究,结合长短期记忆(long short term memory,LSTM)网络和深度残差收缩网络(deep residual shrinkage network,DRSN),设计了时序深度残差收缩网络模型,其中包含残差模块、收缩模块和LSTM模块。残差模块和收缩模块负责提取混叠信号中的显著信息并自适应生成决策阈值,LSTM模块用于提取混叠信号中的时序隐含特征。三者结合可以有效提高混叠信号的识别精度。公开和实测数据集测试结果表明,所提方法识别精度优于5种典型方法,在高信噪比下的平均识别分类准确率可以达到92.7%;21种混叠信号中有12种识别准确率接近100%。展开更多
基金supported by the National Natural Science Foundation of China(61471194)the Fundamental Research Funds for the Central Universities+2 种基金the Science and Technology on Avionics Integration Laboratory and Aeronautical Science Foundation of China(20155552050)the CASC(China Aerospace Science and Technology Corporation) Aerospace Science and Technology Innovation Foundation Projectthe Nanjing University of Aeronautics And Astronautics Graduate School Innovation Base(Laboratory)Open Foundation Program(kfjj20151505)
文摘The traditional oriented FAST and rotated BRIEF（ORB） algorithm has problems of instability and repetition of keypoints and it does not possess scale invariance. In order to deal with these drawbacks, a modified ORB（MORB） algorithm is proposed. In order to improve the precision of matching and tracking, this paper puts forward an MOK algorithm that fuses MORB and Kanade-Lucas-Tomasi（KLT）. By using Kalman, the object＇s state in the next frame is predicted in order to reduce the size of search window and improve the real-time performance of object tracking. The experimental results show that the MOK algorithm can accurately track objects with deformation or with background clutters, exhibiting higher robustness and accuracy on diverse datasets. Also, the MOK algorithm has a good real-time performance with the average frame rate reaching 90.8 fps.
文摘为提高自然环境中玉米害虫识别的准确性,开发一种基于优化YOLOv8的深度学习模型YOLOv8-LAP。该模型将大型可分离卷积核注意力(LSKA)机制引入特征融合模块空间快速金字塔池化(SPPF),增强多尺度特征提取能力,提升检测性能。针对玉米害虫图像检测中小目标难以捕捉、背景复杂和光照变化等挑战,在主干网络中加入AFGC(Attention for Fine-Grained Categorization)层,以进一步增强图像特征提取的效果,提升模型的泛化能力和鲁棒性。为保证实时检测和模型轻量化,引入可编程梯度信息(PGI)技术,通过辅助监督优化训练过程,减少参数并加速推理。在9种常见玉米害虫的检测中,YOLOv8-LAP模型的平均精度均值(mAP0.5)达到了95.7%,相较于原始YOLOv8模型提高了4.9个百分点。此外,为验证YOLOv8-LAP模型的效果,开发一款基于PySide6的应用程序,该应用拥有用户友好的图形用户界面(GUI),具有实时图像处理和视频分析功能,并支持静态图像、动态视频和摄像头实时目标检测。可见,YOLOv8-LAP模型在降低漏检率和误检率方面表现突出,目标定位更精准,适用于自然环境下的玉米害虫识别,并为精准施药提供技术支持。
文摘基于深度学习进行信号自动调制识别在分类精度、可迁移性等方面普遍优于传统方法,引起广泛关注。但是,当前方法多数针对单信号样本进行识别,无法适用于混叠信号识别场景。针对该问题,对混叠信号调制识别方法进行了研究,结合长短期记忆(long short term memory,LSTM)网络和深度残差收缩网络(deep residual shrinkage network,DRSN),设计了时序深度残差收缩网络模型,其中包含残差模块、收缩模块和LSTM模块。残差模块和收缩模块负责提取混叠信号中的显著信息并自适应生成决策阈值,LSTM模块用于提取混叠信号中的时序隐含特征。三者结合可以有效提高混叠信号的识别精度。公开和实测数据集测试结果表明,所提方法识别精度优于5种典型方法,在高信噪比下的平均识别分类准确率可以达到92.7%;21种混叠信号中有12种识别准确率接近100%。
