目标检测技术正逐步应用于农业,然而在农田害虫检测的运用中仍存在检测速度慢、检测准确率偏低的问题,且仅仅预测害虫的种类和位置信息不足以满足复杂的工程需求。提出一种可以额外预测害虫状态信息的融合MAE和YOLOX算法的高速高精度农...目标检测技术正逐步应用于农业,然而在农田害虫检测的运用中仍存在检测速度慢、检测准确率偏低的问题,且仅仅预测害虫的种类和位置信息不足以满足复杂的工程需求。提出一种可以额外预测害虫状态信息的融合MAE和YOLOX算法的高速高精度农田害虫检测模型MRA-YOLOX(masked autoencoders and rapid aim detection-exceeding YOLO)。算法构建包含近4万张图片以及5万余标注的数据集TDBFP(target detection dataset be used for farmland pests),TDBFP数据集标注了10种害虫的生长状态、物种类别以及位置,以便更好地把握害虫信息,从而更准确地制定对策。修改YOLOX模型的解耦头及loss,额外输出生长状态,以改进模型预测更多信息;将ECA(efficient channel attention)和SA(shuffle attention)注意力机制进行有机融合,并插入backbone与FPN(feature pyr-amid networks)的连接过程以及FPN的通道堆叠过程,以便能够增强获得全局信息和丰富上下文信息的能力,从而取得比单一注意力机制更好的效果;将MAE中自监督解码器部分插入YOLOX的数据增强部分,扩大感受野,增强识别细粒度,获得超越mixup和mosaic的数据增强效果。实验结果表明,当需要同时感知目标的状态、分类和位置时,MRA-YOLOX相较于原始YOLOX模型对于TDBFP数据集的检测精度mAP@0.5由60.1%上升到88.2%,平均检测准确率提高了18.8个百分点,且检测帧率达到145 FPS,可以用于更复杂的工程实践。展开更多
1 Introduction Interaction of radio-frequency wave with plasma in magnetic confinement devices has been a very important discipline of plasma physics. To approach more realistic description of wave-plasma interaction...1 Introduction Interaction of radio-frequency wave with plasma in magnetic confinement devices has been a very important discipline of plasma physics. To approach more realistic description of wave-plasma interaction in a time scale longer than the kinetic time scales bounce-average is needed. The long time evolution of the kinetic distribution can be treated by Fokker-Planck equation. The behavior of the plasma and the most interesting macroscopic effects are obtained by balancing the diffusion term with a collision term.展开更多
1 Introduction Internal kink mode driven by barely trapped suprakthermal electrons is reported on D Ⅲ-D tokama . This instability is most active when electron cyclotron current drive （ ECCD ） is applied on the hi...1 Introduction Internal kink mode driven by barely trapped suprakthermal electrons is reported on D Ⅲ-D tokama . This instability is most active when electron cyclotron current drive （ ECCD ） is applied on the high field side of the flux surface. It has a bursting behavior with poloidal/toroidal mode number m/n = 1/1. In positive magnetic shear plasma this mode becomes fishbone instability.展开更多
1 Introduction Internal kink mode driven by the supra-thermal electrons is reported on DⅢ-D tokamak. This instability is most active when electron cyclotron current drive （ECCD） is applied on the high field side o...1 Introduction Internal kink mode driven by the supra-thermal electrons is reported on DⅢ-D tokamak. This instability is most active when electron cyclotron current drive （ECCD） is applied on the high field side of the flux surface. It has a bursting behavior with poloidal/toroidal mode number=n/n=1/1. In positive magnetic shear plasma this mode becomes fishbone instability. The mode frequency is about 10 kHz. Strong m=1 MHD activities are also observed in HL-IM tokamak during off-axis electron cyclotron resonance heating （ECRH） when the cyclotron location is placed just outside the q= 1 flux surface at high field side .展开更多
文摘目标检测技术正逐步应用于农业,然而在农田害虫检测的运用中仍存在检测速度慢、检测准确率偏低的问题,且仅仅预测害虫的种类和位置信息不足以满足复杂的工程需求。提出一种可以额外预测害虫状态信息的融合MAE和YOLOX算法的高速高精度农田害虫检测模型MRA-YOLOX(masked autoencoders and rapid aim detection-exceeding YOLO)。算法构建包含近4万张图片以及5万余标注的数据集TDBFP(target detection dataset be used for farmland pests),TDBFP数据集标注了10种害虫的生长状态、物种类别以及位置,以便更好地把握害虫信息,从而更准确地制定对策。修改YOLOX模型的解耦头及loss,额外输出生长状态,以改进模型预测更多信息;将ECA(efficient channel attention)和SA(shuffle attention)注意力机制进行有机融合,并插入backbone与FPN(feature pyr-amid networks)的连接过程以及FPN的通道堆叠过程,以便能够增强获得全局信息和丰富上下文信息的能力,从而取得比单一注意力机制更好的效果;将MAE中自监督解码器部分插入YOLOX的数据增强部分,扩大感受野,增强识别细粒度,获得超越mixup和mosaic的数据增强效果。实验结果表明,当需要同时感知目标的状态、分类和位置时,MRA-YOLOX相较于原始YOLOX模型对于TDBFP数据集的检测精度mAP@0.5由60.1%上升到88.2%,平均检测准确率提高了18.8个百分点,且检测帧率达到145 FPS,可以用于更复杂的工程实践。
文摘1 Introduction Interaction of radio-frequency wave with plasma in magnetic confinement devices has been a very important discipline of plasma physics. To approach more realistic description of wave-plasma interaction in a time scale longer than the kinetic time scales bounce-average is needed. The long time evolution of the kinetic distribution can be treated by Fokker-Planck equation. The behavior of the plasma and the most interesting macroscopic effects are obtained by balancing the diffusion term with a collision term.
基金Supported by National Natural Science Foundation of China （ 19885006, 10175020 and 10135020 ）
文摘1 Introduction Internal kink mode driven by barely trapped suprakthermal electrons is reported on D Ⅲ-D tokama . This instability is most active when electron cyclotron current drive （ ECCD ） is applied on the high field side of the flux surface. It has a bursting behavior with poloidal/toroidal mode number m/n = 1/1. In positive magnetic shear plasma this mode becomes fishbone instability.
文摘1 Introduction Internal kink mode driven by the supra-thermal electrons is reported on DⅢ-D tokamak. This instability is most active when electron cyclotron current drive （ECCD） is applied on the high field side of the flux surface. It has a bursting behavior with poloidal/toroidal mode number=n/n=1/1. In positive magnetic shear plasma this mode becomes fishbone instability. The mode frequency is about 10 kHz. Strong m=1 MHD activities are also observed in HL-IM tokamak during off-axis electron cyclotron resonance heating （ECRH） when the cyclotron location is placed just outside the q= 1 flux surface at high field side .
