[ Objective] The aim was to compare control effects of trapping lamps with different wavelengths on soybean pests. [ Method ] Three kinds of lamps with different wavelengths were used to conduct pest control test in s...[ Objective] The aim was to compare control effects of trapping lamps with different wavelengths on soybean pests. [ Method ] Three kinds of lamps with different wavelengths were used to conduct pest control test in soybean fields. [ Result] The trapping effects of three kinds of lamps with different wavelengths on soybean pests were significantly different. Among them, the trapping effect of No. 2 light with wavelength of 400 -460 nm was significantly better than the other two lamps; No. 1 light with wavelength of 330 -400 nm had no obvious difference in trapping effect with No. 3 light with wavelength of 450 -510 nm. [ Conclusion] The research result provided theoretical basis for popularized application of trapping lamp in production.展开更多
The trapping effect of Jiaduo frequoscilation pest-killing lamp was studied in vegetable fields in Jianghan plain from 2009 to 2010. The results showed that the insects trapped by the lamp were mainly Lepidoptcra and ...The trapping effect of Jiaduo frequoscilation pest-killing lamp was studied in vegetable fields in Jianghan plain from 2009 to 2010. The results showed that the insects trapped by the lamp were mainly Lepidoptcra and Coleoptera insects. There was no significant difference between the trapped numbers in the two years, but there were significant difference among the months. The peak of species richness was nearly 50 ; the diversity index was from 2.0 to 4.0, and the even- ness index varied from 0. 3 to O. 8, while the dominant concentration index of insects varied from 0.05 to 0. 50. It was helpful to detect and prevent insect pests ac- cording to their peak occurrence time inferred by trapping results.展开更多
田间害虫图像数据采集困难,并且传统的检测模型大多使用复杂的特征金字塔(Feature pyramid network,FPN)结构提升精度,这在一定程度上影响了检测的实时性。为此,本研究通过设计诱虫灯装置构建害虫数据集FieldPest5,并且对无FPN结构的检...田间害虫图像数据采集困难,并且传统的检测模型大多使用复杂的特征金字塔(Feature pyramid network,FPN)结构提升精度,这在一定程度上影响了检测的实时性。为此,本研究通过设计诱虫灯装置构建害虫数据集FieldPest5,并且对无FPN结构的检测器YOLOF进行改进,提出兼顾检测精度和效率的害虫检测模型YOLOF_PD。首先,增加Cutout数据增强方法缓解害虫图像中的遮挡问题,并且使用CIoU损失函数获得更好的框回归位置;其次,在原有坐标注意力机制(Coordinate attention,CA)的全局平均池化(Global average pooling,GAP)路径中增加全局最大池化(Global max pooling,GMP)路径,并且使用可学习参数自适应更新不同路径的权重,提出自适应坐标注意力机制(Adaptive coordinate attention,ACA),增强模型的信息表征能力;最后,对YOLOF膨胀编码器中的Projector和Residual模块进行改进,在Projector模块的3×3卷积后引入ACA注意力机制,在Residual模块中融合3×3的深度可分离卷积和1×1的逐点卷积,提出Dilated_Dwise_ACA编码器,提高YOLOF对小尺度害虫的检测性能。实验结果表明:改进后的YOLOF_PD模型在FieldPest5测试集上的平均精度均值(Mean average precision,mAP)为93.7%,较改进前提升2.1个百分点,并且检测时图像传输速率为42.4 f/s,能够满足害虫快速检测的要求。对比Cascade R CNN、RetinaNet、ATSS等模型,YOLOF_PD模型在检测效果和检测速度方面均取得了良好性能。展开更多
基金Supported by Project of National Soybean Industrial Technology System ( nycy - 004)
文摘[ Objective] The aim was to compare control effects of trapping lamps with different wavelengths on soybean pests. [ Method ] Three kinds of lamps with different wavelengths were used to conduct pest control test in soybean fields. [ Result] The trapping effects of three kinds of lamps with different wavelengths on soybean pests were significantly different. Among them, the trapping effect of No. 2 light with wavelength of 400 -460 nm was significantly better than the other two lamps; No. 1 light with wavelength of 330 -400 nm had no obvious difference in trapping effect with No. 3 light with wavelength of 450 -510 nm. [ Conclusion] The research result provided theoretical basis for popularized application of trapping lamp in production.
基金Supported by Open Fund for Engineering Research Center of Ecology and Agricultural Use of Wetland,Ministry of Education&Fund for Outstanding Young Teachers in Yangtze University
文摘The trapping effect of Jiaduo frequoscilation pest-killing lamp was studied in vegetable fields in Jianghan plain from 2009 to 2010. The results showed that the insects trapped by the lamp were mainly Lepidoptcra and Coleoptera insects. There was no significant difference between the trapped numbers in the two years, but there were significant difference among the months. The peak of species richness was nearly 50 ; the diversity index was from 2.0 to 4.0, and the even- ness index varied from 0. 3 to O. 8, while the dominant concentration index of insects varied from 0.05 to 0. 50. It was helpful to detect and prevent insect pests ac- cording to their peak occurrence time inferred by trapping results.
文摘田间害虫图像数据采集困难,并且传统的检测模型大多使用复杂的特征金字塔(Feature pyramid network,FPN)结构提升精度,这在一定程度上影响了检测的实时性。为此,本研究通过设计诱虫灯装置构建害虫数据集FieldPest5,并且对无FPN结构的检测器YOLOF进行改进,提出兼顾检测精度和效率的害虫检测模型YOLOF_PD。首先,增加Cutout数据增强方法缓解害虫图像中的遮挡问题,并且使用CIoU损失函数获得更好的框回归位置;其次,在原有坐标注意力机制(Coordinate attention,CA)的全局平均池化(Global average pooling,GAP)路径中增加全局最大池化(Global max pooling,GMP)路径,并且使用可学习参数自适应更新不同路径的权重,提出自适应坐标注意力机制(Adaptive coordinate attention,ACA),增强模型的信息表征能力;最后,对YOLOF膨胀编码器中的Projector和Residual模块进行改进,在Projector模块的3×3卷积后引入ACA注意力机制,在Residual模块中融合3×3的深度可分离卷积和1×1的逐点卷积,提出Dilated_Dwise_ACA编码器,提高YOLOF对小尺度害虫的检测性能。实验结果表明:改进后的YOLOF_PD模型在FieldPest5测试集上的平均精度均值(Mean average precision,mAP)为93.7%,较改进前提升2.1个百分点,并且检测时图像传输速率为42.4 f/s,能够满足害虫快速检测的要求。对比Cascade R CNN、RetinaNet、ATSS等模型,YOLOF_PD模型在检测效果和检测速度方面均取得了良好性能。
