基于改进 YOLOv8n的夏玉米土壤背景分割与长势监测方法研究

Background Segmentation and Growth Monitoring Methods for Summer Maize Based on an Improved YOLOv8n Approach

摘要

叶面积指数（Leaf Area Index， LAI）、叶绿素含量（Soil and Plant Analyzer Development， SPAD）、株高（Vegetation Height， VH）是玉米长势监测的重要指标。然而，在基于遥感或近地光谱影像的反演过程中，土壤背景干扰常显著降低这些关键参数的估算精度。近年来，尽管深度学习技术在农业遥感领域得到广泛应用，但在复杂田间场景下，如何有效分离作物冠层与背景噪声、提升特征区分能力，尤其是针对玉米这类形态结构不规则、冠层分布不均匀的作物，仍是当前研究中的难点。本研究针对特征区分能力弱、不规则植株形态影响特征提取等问题，提出了基于改进YOLOv8n的土壤背景去除模型（YOLOv8n-SA-W-DCN）。该模型在YOLOv8n中融合SimAM无参数注意力机制以增强模型对玉米冠层与土壤背景的特征区分能力，引入可变形卷积（Deformable Convolution Network， DCN）以精准拟合不规则植株形态，并利用WIoU损失函数动态优化训练过程。试验结果证明，改进后模型的mAP较原始YOLOv8n提升了1.6%，有效去除了土壤背景的干扰，从而实现夏玉米长势参数的精准监测。在此基础上，本研究基于去除背景后的纯净冠层影像，确定了LAI、SPAD及VH在各生育期的最优反演模型。与基于原始影像的反演模型相比，本方法在LAI、SPAD和VH反演中的R ²进一步提升（2%-11%），MRE及RMSE均明显下降。研究结果表明，提出的土壤背景去除方法有效性高，能实现对夏玉米长势的精准监测，为作物表型分析及精准农业管理提供了技术支撑。

Abstract

Leaf Area Index (LAI), Soil and Plant Analyzer Development (SPAD), and Vegetation Height (VH) are important indicators for monitoring maize growth. However, during the inversion process based on remote sensing or near-ground spectral imagery, soil background interference often significantly reduces the estimation accuracy of these key parameters. In recent years, despite the widespread application of deep learning technologies in agricultural remote sensing, effectively separating crop canopies from background noise and enhancing feature discrimination capabilities in complex field scenarios -particularly for crops like maize with irregular morphological structures and uneven canopy distributions- remains a major challenge in current research. This study addresses issues such as weak feature discriminative ability and the impact of irregular plant morphology on feature extraction by proposing an improved YOLOv8n-based soil background removal model (YOLOv8n-SA-W-DCN). The model integrates a parameter-free SimAM attention mechanism into YOLOv8n to enhance the feature differentiation between maize canopy and soil background. It also incorporates a Deformable Convolution Network (DCN) to accurately adapt to irregular plant shapes, and employs a WIoU loss function to dynamically optimize the training process. Experimental results show that the improved model achieves a 1.6% increase in mean average precision (mAP) compared to the original YOLOv8n, effectively removing soil background interference and enabling precise monitoring of summer maize growth parameters. Furthermore, based on the cleaned canopy images after background removal, this study establishes optimal inversion models for LAI, SPAD, and VH at different growth stages. Compared with models based on original images, the proposed method further improves the coefficient of determination (R ²) by 2%–11% in the inversion of LAI, SPAD, and VH, while significantly reducing the Mean Relative Error (MRE) and Root Mean Square Error (RMSE). These results demonstrate the high effectiveness of the proposed soil background removal method in achieving accurate monitoring of summer maize growth, providing reliable technical support for crop phenotyping analysis and precision agriculture management.

关键词

图像分割 / 无人机 / 注意力模型 / 多光谱影像 / 作物长势监测 / 夏玉米

Key words

Image segmentation / unmanned aerial vehicles / attention models / multispectral imagery / crop growth monitoring / summer maize

参考文献

[1] 仇焕广, 李新海, 余嘉玲. 中国玉米产业:发展趋势与政策建议[J]. 农业经济问题, 2021(07): 4-16.

[2] 李少昆, 赵久然, 董树亭, 等. 中国玉米栽培研究进展与展望[J]. 中国农业科学, 2017, 50(11): 1941-1959.

[3] 王鹏新, 杜江莉, 张悦, 等. 基于遥感多参数和CNN-Transformer的冬小麦单产估测[J]. 农业机械学报, 2024, 55(03): 173-182.

[4] 赵经华, 马世骄, 房城泰. 基于熵权-模糊综合评价法的无人机多光谱春玉米长势监测模型研究[J]. 农业机械学报, 2024, 55(08): 214-224.

[5] Sun X, Zhang Y, Shi K, et al. Monitoring water quality using proximal remote sensing technology [J]. Science of the Total Environment, 2022, 803: 149805.

[6] Olson D, Anders O J . Review on unmanned aerial vehicles, remote sensors, imagery processing, and their applications in agriculture [J]. Agronomy Journal, 2021, 113(2): 971-992.

[7] 许童羽, 白驹驰, 郭忠辉, 等. 基于无人机高光谱遥感的水稻氮营养诊断方法[J]. 农业机械学报, 2023, 54(02): 189-197,222.

[8] 高姻燕, 孙义, 李葆春. 基于无人机RGB影像估测田间小麦穗数[J]. 中国农业科技导报, 2022, 24(03): 103-110.

[9] 郑超磊, 贾立, 胡光成. 高分一号卫星遥感数据驱动ETMonitor模型估算16m分辨率蒸散发及验证[J]. 遥感学报, 2023, 27(03): 758-768.

[10] 傅友强, 钟旭华, 黄农荣, 等. 基于无人机多光谱遥感的水稻冠层光谱特征和氮素营养关系研究[J]. 广东农业科学, 2021, 48(10): 121-131.

[11] 王佳丽, 蒯雁, 杨成伟, 等. 基于无人机多光谱的烤烟冠层叶绿素含量反演[J]. 江苏农业科学, 2024, 52(15): 232-238.

[12] 王伟康, 张嘉懿, 汪慧, 等. 基于固定翼无人机多光谱影像的水稻长势关键指标无损监测[J]. 中国农业科学, 2023, 56(21): 4175-4191.

[13] 邵亚杰, 汤秋香, 崔建平, 等. 融合无人机光谱信息与纹理特征的棉花叶面积指数估测[J]. 农业机械学报, 2023, 54(06): 186-196.

[14] Liu S, Jin X, Nie C, et al. Estimating leaf area index using unmanned aerial vehicle data: shallow vs deep machine learning algorithms[J]. Plant Physiology, 2021, 187(3): 1551-1576.

[15] 罗小波, 谢天授, 董圣贤. 基于无人机多光谱影像的柑橘冠层叶绿素含量反演[J]. 农业机械学报, 2023, 54(04): 198-205.

[16] 王晗, 向友珍, 李汪洋, 等. 基于无人机多光谱遥感的冬油菜地上部生物量估算[J]. 农业机械学报, 2023, 54(08): 218-229.

[17] 高睿, 吴会才, 刘春菊, 等. 基于改进YOLOv8的烟草植株检测与计数方法研究[J]. 中国烟草学报, 2025, 31(01): 111-122.

[18] 张蔚然, 杜岳峰, 栗晓宇, 等. 基于FSLYOLO v8n的玉米籽粒收获质量在线检测方法研究[J]. 农业机械学报, 2024, 55(08): 253-265.

[19] 孙刚, 黄文江, 陈鹏飞, 等. 轻小型无人机多光谱遥感技术应用进展[J]. 农业机械学报, 2018, 49(03): 1-17.

[20] 高睿, 荆茹彬, 刘春菊, 等. 基于改进YOLOv8s模型的烟草5种常见病害的智能检测[J]. 烟草科技, 2025, 58(11): 33-43.

[21] Yang L, Zhang R Y, Li L, et al. Simam: A simple, parameter-free attention module for convolutional neural networks[C]. International Conference on Machine Learning. PMLR, 2021: 11863-11874.

[22] Cho Y J. Weighted Intersection over Union (wIoU) for evaluating image segmentation[J]. Pattern Recognition Letters, 2024, 185: 101-107.

[23] Dai J, Qi H, Xiong Y, et al. Deformable convolutional networks[C]. Proceedings of the IEEE International Conference on Computer Vision. 2017: 764-773.

[24] Xue J, Su B. Significant remote sensing vegetation indices: A review of developments and applications [J]. Journal of Sensors, 2017, 2017(1): 1353691.

[25] Dash J, Jeganathan C, Atkinson P M. The use of MERIS terrestrial chlorophyll index to study spatio-temporal variation in vegetation phenology over India[J]. Remote Sensing of Environment, 2010, 114(7): 1388-1402.

[26] Roujean J L, Breon F M. Estimating PAR absorbed by vegetation from bidirectional reflectance measurements[J]. Remote Sensing of Environment, 1995, 51(3): 375-384.

[27] 付新阳, 崔利华, 董雨昕, 等. 去除土壤背景影响的多光谱遥感影像玉米叶面积指数估算[J]. 农业机械学报, 2025, 56(05): 384-394.

[28] 孙磊康, 李孝永, 郭航兆, 等. 施氮量和种植密度对玉米冠层光截获、籽粒灌浆和产量的影响[J]. 农业工程学报, 2025, 41(22): 101-113.

[29] 吴婷婷, 刘昕哲, 聂睿琪, 等. 基于细粒度校正的育种小区小麦株高无人机测量方法[J]. 农业机械学报, 2023, 54(06): 158-167.

基金资助

山东省自然科学基金项目(ZR2023MC201)

山东省重大科技创新工程项目(2019JZZY010716)

国家重点研发计划项目(2023YFD200140403)