首页|基于NanoDet-SimAM小尺寸松材线虫病受害木检测

基于NanoDet-SimAM小尺寸松材线虫病受害木检测

扫码查看
原文链接
  • 万方数据
  • 维普
针对小尺寸松材线虫病受害木检测精度及检测效率低的问题,提出了一种融合深度网络和注意力机制的小尺寸松材线虫智能检测模型.采用无人机(UAV)搭载小型相机在 220 m高度拍摄小尺寸松材线虫受害木图像,应用图像旋转、缩放、添加高斯噪声和模拟光照强度等数据处理方式扩充数据集,设计轻量级深度网络NanoDet和SimAM注意力模块融合模型NanoDet-SimAM对小尺寸松材线虫受害木进行精准检测.结果表明,该模型相较于Faster R-CNN、Yolov4、Yolov5s及NanoDet等检测网络模型,具有更高的检测精度、速度和稳定性.
Detection based on NanoDet-SimAM for small-size trees with pine wilt disease
Aiming at the problems of lower precision and efficiency in the detection of small-size pine wilt disease(PWD)injured trees,an intelligent detection model of small-size PWD injured trees was proposed.This model combined depth network and attention mechanism.A small camera equipped with an unmanned aerial vehicle(UAV)was used to capture images of small-size PWD injured tree at the height of 220 m.Data processing methods were applied to expand the data set.The processing methods included image rotation,scaling,adding Gaussian noise and simulating light intensity.A lightweight deep network NanoDet and a SimAM attention module fusion model NanoDet-SimAM were designed to realize accurate detection of small-size PWD injured trees.The results show that the model has higher detection accuracy,speed and stability than Faster R-CNN,Yolov4,Yolov5s and NanoDet,etc.

pine wilt diseasetarget detectionlightweight network NanoDetattention mechanismnon-parametric attentiontransfer learningdata enhancementsmall size

刘芳、姜生伟、张峻豪、何姗

展开 >

沈阳理工大学理学院,辽宁沈阳 110159

沈阳理工大学辽宁省兵器工业智能优化与控制重点实验室,辽宁沈阳 110159

辽宁省林业和草原局东北林草危险性有害生物防控国家林草局重点实验室,辽宁沈阳 110036

沈阳理工大学自动化与电气工程学院,辽宁沈阳 110159

沈阳工学院东北林草危险性有害生物防控国家林草局重点实验室,辽宁 沈阳 113122

展开 >

松材线虫病 目标检测 轻量级网络NanoDet 注意力机制 无参注意力 迁移学习 数据增强 小尺寸

辽宁省民生科技计划项目辽宁省"兴辽人才"项目辽宁省教育厅基本科研项目

2021JH2/10200008XLYC2006017LJKMZ20220619

2024

10.7688/j.issn.1000-1646.2024.04.11

沈阳工业大学学报
沈阳工业大学

沈阳工业大学学报

CSTPCD北大核心
影响因子：0.62
ISSN：1000-1646
年,卷(期)：2024.46(4)
  • 5