基于ABC-CatBoost和光谱指数的黄瓜霜霉病叶部病斑SPAD值反演与严重度估计

SPAD Value Inversion and Severity Estimation for Cucumber Downy Mildew Leaf Lesions Based on ABC-CatBoost and Spectral Indices

摘要:

准确反演和估计黄瓜霜霉病叶绿素相对含量（SPAD值）和严重度对霜霉病管理具有重要意义。目前相关研究更多着眼于叶片全局，而忽略了局部单个病斑SPAD值反演及严重度估计，且传统SPAD值反演及严重度估计费时费力，本文基于人工蜂群算法优化的分类梯度提升树（ABC-CatBoost）及光谱指数进行黄瓜霜霉病叶部局部单个病斑SPAD值反演及严重度估计研究。通过获取发病叶片最大光化学效率（Fv/Fm）图像，实现不同病斑严重度划分，同时采集相应病斑SPAD值及高光谱数据，并利用最大最小归一化（MMS）及去趋势化（Detrend）数据预处理方法增强光谱数据与SPAD值及严重度的相关性，降低噪声对建模精度的影响。基于构建的一维、二维和三维光谱指数提取了原始、MMS及Detrend预处理下最优光谱特征波段，并基于ABC-CatBoost实现了黄瓜霜霉病叶部病斑SPAD值反演及严重度估计。试验结果表明，ABC-CatBoost对SPAD值反演及严重度估计精度均得到了进一步提升。较一维光谱指数，二维、三维光谱指数凭借多维度信息融合能力，在SPAD值反演中，经Detrend预处理后二维和三维光谱指数与SPAD值决定系数均达到0.927。在严重度估计中，经Detrend预处理后二维光谱指数及经MMS预处理后三维光谱指数与严重度的准确率均达到90.90%，进一步表明Detrend数据预处理方法优于MMS。研究结果不仅为霜霉病管理提供了可靠科学依据，且对提升黄瓜生产产量和品质具有重要意义。

Abstract:

Accurate inversion and estimation of SPAD content and severity of cucumber downy mildew are of great significance for the management of the disease.Current research predominantly focused on overall leaf analysis, overlooking the inversion and estimation of SPAD content and severity of individual lesions.Moreover, traditional methods for SPAD inversion and severity estimation were time-consuming and labor-intensive.Therefore, it was aimed to investigate the SPAD inversion and severity estimation of individual cucumber downy mildew lesions based on ABC-CatBoost and spectral indices.By acquiring Fv/Fm images of infected leaves, the severity of different lesions was classified.Corresponding SPAD values and hyperspectral data of the lesions were also collected.To enhance the correlation between spectral data and SPAD content and severity, and reduce the impact of noise on modeling accuracy, MMS and Detrend data preprocessing methods were utilized.Optimal spectral feature bands were extracted from raw, MMS, and Detrend preprocessed data based on constructed one-dimensional, two-dimensional, and three-dimensional spectral indices.Using these features, the ABC-CatBoost model was employed to achieve SPAD inversion and severity estimation of cucumber downy mildew lesions.The results showed that the CatBoost algorithm, optimized by the artificial bee colony (ABC) algorithm, significantly improved the accuracy of SPAD inversion and severity estimation.Compared with one-dimensional spectral indices, two-dimensional and three-dimensional spectral indices, with their ability to integrate multidimensional information, demonstrated superior performance.Specifically, after Detrend preprocessing, the coefficients of determination between the two-dimensional and three-dimensional spectral indices and SPAD reached 0.927.In severity estimation, the accuracy between the Detrend-preprocessed two-dimensional spectral indices and the MMS-preprocessed three-dimensional spectral indices and the severity reached 90.90%, further demonstrating that Detrend preprocessing was superior to MMS.The research result can not only provide a reliable scientific basis for managing downy mildew but also have significant implications for improving the yield and quality of cucumber production.