PMJDM：一种用于植物病害识别的多任务联合检测模型。
PMJDM: a multi-task joint detection model for plant disease identification.
作者信息
Fu Rui, Wang Xuewei, Wang Shiyu, Sun Hao
机构信息
Shandong Province University Laboratory for Protected Horticulture, Weifang University of Science and Technology, Weifang, China.
出版信息
Front Plant Sci. 2025 May 22;16:1599671. doi: 10.3389/fpls.2025.1599671. eCollection 2025.
INTRODUCTION
Plant disease detection is critical for ensuring agricultural productivity, yet traditional methods often suffer from inefficiencies and inaccuracies due to manual processes and limited adaptability.
METHODS
This paper presents the PlantDisease Multi-task Joint Detection Model (PMJDM), which integrates an enhanced ConvNeXt-based shared feature extraction, a texture-augmented N-RPN module with HOG/LBP metrics, multi-task branches for simultaneous plant species classification and disease detection, and CRF-based post-processing for spatial consistency. A dynamic weight adjustment mechanism is also employed to optimize task balance and improve robustness.
RESULTS
Evaluated on a 26,073-image dataset, PMJDM achieves 71.84% precision, 61.96% recall, and 61.83% mAP50, surpassing Faster - RCNN (51.49% mAP50) and YOLOv10x (59.52% mAP50) by 10.34% and 2.31%, respectively.
DISCUSSION
The superior performance of PMJDM is driven by multi-task synergy and texture - enhanced region proposals, offering an efficient solution for precision agriculture.
引言
植物病害检测对于确保农业生产力至关重要，但传统方法由于人工操作流程和有限的适应性，往往效率低下且不准确。
方法
本文提出了植物病害多任务联合检测模型（PMJDM），该模型集成了基于增强型ConvNeXt的共享特征提取、具有HOG/LBP度量的纹理增强型N-RPN模块、用于同时进行植物物种分类和病害检测的多任务分支，以及基于CRF的后处理以实现空间一致性。还采用了动态权重调整机制来优化任务平衡并提高鲁棒性。
结果
在一个包含26,073张图像的数据集上进行评估时，PMJDM的精确率达到71.84%，召回率达到61.96%，mAP50达到61.83%，分别比Faster - RCNN（mAP50为51.49%）和YOLOv10x（mAP50为59.52%）高出10.34%和2.31%。
讨论
PMJDM的卓越性能得益于多任务协同和纹理增强的区域提议，为精准农业提供了一种高效的解决方案。
Zotero 插件