Real-time identification and spatial distribution mapping of weeds through unmanned aerial vehicle (UAV) remote sensing

文献类型： 外文期刊

第一作者： Luo, Weisong

作者： Luo, Weisong;Chen, Qifan;Wang, Yubo;Fu, Di;Mi, Zhiwen;Wang, Qifan;Su, Baofeng;Luo, Weisong;Chen, Qifan;Wang, Yubo;Fu, Di;Mi, Zhiwen;Wang, Qifan;Su, Baofeng;Luo, Weisong;Chen, Qifan;Wang, Yubo;Fu, Di;Mi, Zhiwen;Wang, Qifan;Su, Baofeng;Li, Huibin;Shi, Yun;Li, Huibin;Shi, Yun

作者机构：

关键词： High-precision positioning; Real-time weed species identification; SHATL-YOLOV8s; UAV video data; Weed spatial distribution maps

期刊名称：EUROPEAN JOURNAL OF AGRONOMY （ 影响因子：5.5； 五年影响因子：5.9 ）

ISSN： 1161-0301

年卷期： 2025 年 169 卷

页码：

收录情况： SCI

摘要： Accurate and efficient weed identification and spatial distribution mapping are essential for precise variable herbicide spraying and optimization of weed management strategies. This study proposes a novel approach for real-time weed identification and spatial distribution mapping using unmanned aerial vehicle (UAV) remote sensing. The SHATL-YOLOV8s model and UAV video data were used to identify weed species in wheat fields. High-precision positioning technology was employed to locate weeds and generate weed spatial distribution maps. The Slide loss function was incorporated into the YOLOv8s model to deal with imbalanced samples and improve model accuracy. The Haar wavelet downsampling (HWD)-Adown module was used instead of the convolution (Conv) module for downsampling, increasing the number of feature channels. The Task Dynamic Align Detection Head (TDADH) was designed to improve capturing weed features. The model was pruned using the Layer-adaptive Magnitude-based Pruning (LAMP) algorithm, significantly reducing its number of parameters (Params) and floating-point operations (FLOPs). The results showed that SHATL-YOLOV8s performed well in weed identification in complex environments, achieving a mean average precision (mAP) of 89.10 %, a precision (P) of 87.80 %, a recall (R) of 82.72 %, and an F1-score of 85.19 %. These values were 3.90 %, 3.56 %, 1.68 %, and 2.58 % higher than those of YOLOv8s, respectively. In addition, the Params was 3.10 M (72.15 % lower than YOLOv8s'), the FLOPs were 16.00 G (43.66 % lower), the model size was 6.20 MB (71.16 % lower), and the detection speed was 80.64 frames per second (FPS) (7.18 % faster). The SHATL-YOLOV8s had fewer parameters and FLOPs and a faster detection speed than the classic models. Tests on Android devices demonstrated that the identification accuracy exceeded 88.00 %, and the detection speed was 29.31 FPS, meeting real-time detection requirements. The weeds' latitude and longitude coordinates were obtained by the UAV's high-precision positioning technology. Weed spatial distribution maps with three categories had an average precision of 89.98 %. The proposed method provides valuable technical guidance for variable pesticide application in precision agriculture and promoting sustainable agricultural development.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序