基于YOLOv5的复杂背景下植物叶片检测研究

Plant Leaf Detection Research in Complex Backgrounds Based on YOLOv5

对植物叶片进行检测是研究植物表型性状的基础,但真实环境下叶片间相互遮挡、叶片边缘特征不明显、幼叶目标过小以及外部环境如光照条件等因素影响会对叶片检测效果造成很大的障碍。针对复杂背景下的叶片检测,该研究提出了一种基于改进YOLOv5模型植物叶片检测方法。通过在骨干网络中引入空洞卷积,使得网络可以捕获到更广阔范围的上下文信息;利用双向连接的加权特征金字塔网络,以增强目标叶片特征提取并更好地融合特征信息;利用注意力机制,通过动态地调整注意力分布,以提高边缘特征表达能力。测试结果表明,在Plant Village数据集筛选的葡萄叶片图像以及自拍摄葡萄生长叶片上测试改进算法的可行性,改进的YOLOv5模型其叶片检测mAP比原生模型提高了5.8%,遮挡叶片检测精度提高了7.09%。叶片检测效果有显著提升。该研究提出的方法可以有效解决复杂背景下植物叶片检测效果不佳的问题,为植物表型研究提供技术支撑。

Detecting plant leaves forms the basis for studying plant phenotypic traits.However,real-world conditions,such as mutual leaf occlusion,indistinct leaf edge characteristics,small target leaves,and external factors like varying lighting conditions,pose substantial obstacles to effective leaf detection.To address leaf detection challenges in complex backgrounds,we present an improved plant leaf detection approach based on the YOLOv5 model.The introduction of dilated convolutions in the backbone network extends the network's ability to capture a wider contextual information range.Leveraging a bidirectional connected and weighted feature pyramid network enhances the extraction of target leaf features and improves feature information integration.The incorporation of attention mechanisms dynamically adjusts attention distributions,thereby enhancing edge feature representation.Test results confirm the feasibility of the enhanced algorithm on grape leaf images from the Plant Village dataset and self-captured grapevine leaf images.The improved YOLOv5 model achieves a 5.8%increase in leaf detection mean average precision(mAP)compared to the original model,with a 7.09%enhancement in occluded leaf detection accuracy.This study significantly enhances leaf detection performance in complex backgrounds.The proposed method effectively addresses the issue of subpar plant leaf detection in such scenarios,thereby providing valuable technical support for plant phenotypic research.