基于无人机高光谱影像的花生苗期株数监测

English title Estimation of the quantity of Peanut seedling plant based on UAV hyperspectral imagery

关键词:  无人机  花生  高光谱成像  群体特征  株数  覆盖度

Key words:  UAV  Peanut  Hyperspectral Imaging  Group characteristics  Number of plants  Coverage

摘要:

【目的】花生是我国重要的油料作物之一,从无人机遥感影像中准确识别花生植株,实现快速、无损、准确的花生株数统计,对花生田间管理和高产具有重大意义。【方法】本研究以广东省河源市灯塔盆地花生种植基地为研究对象,利用无人机获取花生种植区域的高光谱影像数据,利用提取花生平均覆盖度的方法近似估算花生株数。通过分析390～1000nm范围内高光谱图像(HSI)特征,选取分割提取花生植株的有效特征IK波段,构建了包括光谱特征(VIs)和群体形态特征(MPs)在内共20个特征参数,并利用皮尔逊相关分析选择了12个参数。然后基于偏最小二乘回归(Partial Least Square Regression,PLSR)、随机森林回归(Random Forest Regression,RFR)和支持向量回归(Support Vector Regression ,SVR)三种算法,分别利用光谱特征、群体形态特征、联合光谱特征和群体形态特征构建花生平均覆盖度预测模型,并确定估算花生平均覆盖度的最优预测模型。【结果】结果表明,在可见光区域,绿色植株与背景区域的差异明显且不存在重叠区域,671nm处光谱反射率可有效区分花生植株与背景区域。基于选定的特征构建花生株数监测模型,包括 PLSR、RFR 和 SVR。其中在单特征模型中,三种模型效果的优劣次序依次均为群体形态特征指数和最优光谱指数,其中以形态特征指数估测精度最高,R2分别为0.63、0.61和0.67。在多特征模型中,联合光谱特征和群体形态特征与SVR相结合所构建的模型精度更高、稳定性更强,能有效监测田间条件下的花生覆盖度水平,预测的R2和RMSE分别达到0.75和0.29。【结论】研究结果适用于基于无人机的花生苗期株树的实时快速检测,可为花生的精准管理生产提供一种快捷高效的技术手段。

Abstract:

[Purpose] Peanuts stand as a cornerstone of China"s oil crop production. The ability to accurately identify individual peanut plants within a field and obtain precise plant counts is essential for optimizing field management practices and ultimately enhancing yield. Traditional manual counting methods are time-consuming, labor-intensive, and prone to errors. This study aimed to develop and evaluate a robust and efficient method for estimating peanut plant counts leveraging the power of Unmanned Aerial Vehicle (UAV)-based hyperspectral imagery. [Method] This research was conducted in a peanut planting base located in the Dengta Basin of Heyuan City, Guangdong Province, China. High-resolution hyperspectral images (HSI) covering the peanut fields were acquired using a UAV platform. Recognizing the challenges of directly identifying individual plants within dense canopies, we developed a novel approach to estimate plant counts by predicting the average peanut coverage within the canopy. To achieve this, we first delved into analyzing the spectral characteristics of peanut plants across the 390-1000nm spectral range. This analysis aimed to pinpoint specific spectral bands that effectively differentiated peanut plants from the surrounding background, aiding in the segmentation of peanut plants from the imagery. A comprehensive set of 20 features was then meticulously extracted from the hyperspectral data. These features encompassed both spectral indices (VIs), which capture unique spectral reflectance patterns of vegetation, and morphological parameters (MPs), which quantify the shape and structural characteristics of plant canopies. To streamline model development and focus on the most informative features, Pearson correlation analysis was employed to select the 12 most relevant features from the initial set of 20. With the selected features in hand, three powerful machine learning algorithms, namely Partial Least Square Regression (PLSR), Random Forest Regression (RFR), and Support Vector Regression (SVR), were trained and rigorously evaluated to predict the average peanut coverage, which could then be translated into plant count estimates. Each algorithm was trained and evaluated using three distinct feature combinations: spectral indices alone, morphological parameters alone, and a combined set of both spectral and morphological features. This comprehensive approach allowed us to assess the individual contributions of spectral and morphological features, as well as their synergistic potential in predicting peanut coverage. [Result] Analysis of the hyperspectral data revealed distinct spectral signatures between green peanut plants and the surrounding background areas within the visible spectrum. A clear separation was observed at the 671nm wavelength, highlighting its potential as a key feature for accurate peanut plant segmentation in future studies. Among the models trained using only a single feature type, those based on morphological parameters consistently outperformed models relying solely on spectral indices, irrespective of the chosen algorithm (PLSR, RFR, or SVR). This suggests that the spatial arrangement and structural features of peanut canopies, as captured by the morphological parameters, may hold more valuable information for estimating peanut plant counts compared to spectral reflectance patterns alone. The SVR model, utilizing only morphological parameters, achieved the highest predictive accuracy among the single-feature models, boasting an impressive R2 of 0.67. When both spectral and morphological features were combined, the SVR model continued to shine, achieving even higher accuracy and robustness in predicting peanut coverage under real-world field conditions. This powerful model achieved an R2 of 0.75 and an RMSE of 0.29, demonstrating its potential as a valuable tool for real-time peanut plant count monitoring. [Conclusion] This study provides compelling evidence for the effectiveness of UAV-based hyperspectral imagery as a rapid, accurate, and non-destructive tool for peanut plant counting, particularly during the crucial early growth stages. Our findings demonstrate that incorporating both spectral and morphological features into the prediction model significantly improves accuracy. Notably, the SVR model, leveraging both feature types, emerges as a highly promising candidate for real-time monitoring of peanut plants, paving the way for precision agriculture practices in peanut production. By providing timely and accurate plant count estimates, this approach can guide management decisions related to fertilization, irrigation, and pest control, ultimately contributing to increased peanut yield and enhanced resource use efficiency.