基于高光谱成像的番茄叶霉病潜育期检测方法

Detection of Tomato Leaf Mould Latency Based on Hyperspectral Imaging

摘要:

番茄叶霉病的快速传播会导致重大损失，及时检测识别病害具有重要意义。本研究基于高光谱成像技术，标注并提取了170个番茄叶霉病潜育期第5天样本曲线和170个番茄健康样本曲线。通过最大最小归一化（Min-max normalization,MMN）、标准正态变换（Standard normal variate,SNV）、小波变换（Wavelet transform, WT）和基线校正（Baseline correction,BC）4种方法对数据做光谱预处理，使用聚类算法（K-means）剔除异常样本。采用竞争自适应重加权采样（Competitive adaptive reweighted sampling,CARS）算法进行特征波长选择，并对选择出的单特征波长进行定性、定量分析。最后应用支持向量机（Support vector machine,SVM）和线性判别分析（Linear discriminant analysis,LDA）对叶霉病潜育期样本和健康样本进行识别，共构建了8种基于机器学习的番茄叶霉病潜育期检测识别模型并进行对比寻优。结果表明，WT-CARS-LDA模型表现最佳，其识别准确率达到97.62%，为早期发现和防治番茄叶霉病提供了可行的技术方案。

Abstract:

The rapid spread of tomato leaf mould can lead to significant losses, and timely detection and identification of the disease is of great importance. Totally 170 tomato leaf mould latent stage day 5 sample curves and 170 tomato healthy sample curves were labeled and extracted based on hyperspectral imaging. Spectral pre-processing of the data was done by four methods: min-max normalization (MMN), standard normal variate (SNV), wavelet transform (WT) and baseline correction (BC), and abnormal samples were rejected by using a clustering algorithm (K-means). The competitive adaptive reweighted sampling (CARS) algorithm was used for feature band selection, and the selected single feature bands were analyzed qualitatively and quantitatively. Finally, two classification methods, namely support vector machine (SVM) and linear discriminant analysis (LDA), were employed to identify leaf mould latent stage samples and healthy samples, and a total of eight machine learning-based identification models for tomato leaf mould latent stage detection were constructed and compared to find the optimal model. The results showed that the feature bands selected by the CARS algorithm had a positive effect on the overall recognition, and the WT-CARS-LDA model performed the best, with an accuracy of 97.62%. The hyperspectral imaging technology was combined with the machine learning method, and a highly efficient and accurate identification model was successfully constructed for tomato leaf mould potential stage detection. The research result can provide a feasible technical solution for early detection and control of tomato leaf mould.