摘要
针对复杂大田场景中紫花苜蓿倒伏区域边界模糊、形状不规则及小范围倒伏难以准确识别的问题,本研究提出一种基于无人机RGB(R、G、B分别表示红、绿、蓝)图像的紫花苜蓿倒伏识别方法——AlodgeNet模型。为提升模型对不规则形状与小面积倒伏特征的捕捉能力,并增强空间层次结构学习,在YOLO v8x-seg网络中引入大型可分离卷积核注意力(LSKA)机制和空间深度转化卷积(SPD-Conv),以替换原网络中的部分卷积层。同时在宁夏引黄灌区,通过无人机采集了不同飞行高度(5.0 m、7.5 m、10.0 m)与生育期的紫花苜蓿倒伏RGB图像,并以此构建数据集对模型进行训练。试验结果表明,AlodgeNet模型对飞行高度10.0 m采集图像中紫花苜蓿倒伏区域的识别效果最好,且其对初花期采集图像中紫花苜蓿倒伏区域的识别性能高于分枝期。AlodgeNet模型精确率、召回率、交并比(IoU)阈值为0.50时的平均精度均值(mAP 50)和交并比(IoU)阈值从0.50到0.95(步长0.05)的平均精度均值(mAP 50∶95)分别达到84.9%、79.2%、83.8%和56.7%,整体性能优于YOLO v5x-seg模型、YOLO v10x-seg模型、YOLO v11x-seg模型、YOLO v8x-seg模型、RT-DETR模型和MASK-RCNN模型。相较于原始模型YOLO v8x-seg,AlodgeNet模型mAP 50和mAP 50∶95分别提升5.8个百分点和7.3个百分点。本研究结果为复杂大田环境下紫花苜蓿倒伏的快速识别与面积估算提供了一种高效、便捷的监测手段,有助于实现精准农业中的倒伏灾情评估与管理决策支持。
Aiming at the problems of blurry boundaries,irregular shapes,and difficulty in accurately identifying small-scale lodging areas of Medicago sativa in complex field scenarios,this study proposed a lodging identification method for Medicago sativa based on unmanned air vehicle(UAV)RGB images-the AlodgeNet model.To enhance the model’s ability to capture features of irregularly shaped and small-area lodging regions,as well as strengthen the learning of spatial hierarchical structures,the Large Separable Kernel Attention(LSKA)mechanism and Spatial Pyramid Dilated Convolution(SPD-Conv)were introduced into the YOLOv8x-seg network to replace some of the convolution layers in the original network.Meanwhile,in the Yellow River Irrigation Area of Ningxia,RGB images of Medicago sativa lodging at different flight altitudes(5.0 m,7.5 m,10.0 m)and growth stages were collected by UAV,and a dataset was constructed based on these images for model training.The experimental results showed that the AlodgeNet model achieved the best performance in identifying Medicago sativa lodging regions in images collected at a flight altitude of 10.0 m,and its recognition performance for lodging regions in images collected at the early flowering stage was better than that at the branching stage.The precision,recall,mean average precision at the intersection over union(IoU)threshold of 0.50(mAP 50),and mean average precision at IoU thresholds ranging from 0.50 to 0.95 with a step size of 0.05(mAP 50∶95)of the AlodgeNet model reached 84.9%,79.2%,83.8%,and 56.7%,respectively.Its overall performance outperformed the YOLO v5x-seg,YOLO v10x-seg,YOLO v11x-seg,YOLO v8x-seg,RT-DETR,and MASK-RCNN models.Compared with the original YOLO v8x-seg model,the mAP 50 and mAP 50∶95 of the AlodgeNet model were improved by 5.8 percentage points and 7.3 percentage points,respectively.This study provides an efficient and convenient monitoring method for the rapid identification and area estimation of Medicago sativa lodging in complex field environments,which is conducive to realizing lodging disaster assessment and supporting management decision-making in precision agriculture.
作者
葛永琪
唐道统
刘瑞
朱子欣
李昂
GE Yongqi;TANG Daotong;LIU Rui;ZHU Zixin;LI Ang(School of Information Engineering,Ningxia University,Yinchuan 750021,China;Ningxia Key Laboratory of Artificial Intelligence and Information Security for Channeling Computing Resources from the East to the West,Yinchuan 750021,China)
出处
《江苏农业学报》
北大核心
2026年第1期90-98,共9页
Jiangsu Journal of Agricultural Sciences
基金
国家自然科学基金地区项目(62162052、62262052)。
