摘要
目的 番茄是世界范围内产量高、种植面积广的经济作物之一,具有较高的经济效益。番茄早疫病感染速度快、传播范围广、破坏性强,导致番茄减产减收带来严重的经济损失。番茄早疫病早期病状的判别可以为病害的科学防治提供参考,实现病害区域的可视化定位,精准指导病害防治。本文使用高光谱图像结合机器学习算法建立了番茄早疫病早期病状的判别模型,实现了番茄早疫病早期判别和病害区域的可视化定位。方法 针对高光谱图像的噪声干扰,利用鲁棒主成分分析(RPCA)实现高光谱图像的去噪处理;为了避免主观选择感兴趣区域造成信息表征不足的问题,使用最大类间方差法(OTSU)算法从背景中有效提取了番茄叶片,并以整个番茄叶片的平均光谱作为研究对象,结合多元散射校正(MSC)算法以及标准化建立了光谱预处理的综合方法;针对预处理后的光谱,为了提高模型的判别能力,建立了基于特征波长和线性核函数SVM的番茄早疫病早期判别模型。结果 较全光谱波长建模在测试集上的准确率提高8.33%;对SVM模型的参数进行优化,当参数C=1.64时,训练集和测试集上准确率分别达到了91.67%和94.44%,较参数优化前的模型在训练集上的准确率提高了1.19%,而测试集准确率保持不变,有效缓解了模型欠拟合问题。结论 本文建立了番茄早疫病高光谱图像早期判别模型并实现了早期病状的可视化定位,基于特征波长和线性核函数SVM的判别模型可以有效判别番茄早疫病病害的早期病状;利用患病概率的形式对番茄早疫病早期病状进行可视化定位,可视化分析更直观地发现早期病害并采取防治措施。本研究为植物病害早期病状的判别和可视化研究提供了参考,对作物病害的监测识别和科学防治具有重要意义。
Objective Tomatoes are one of the highest-yielding and most widely cultivated economic crops globally,playing a crucial role in agricultural production and providing significant economic benefits to farmers and related industries.However,early blight in tomatoes is known for its rapid infection,widespread transmission,and severe destructiveness,which significantly impacts both the yield and quality of tomatoes,leading to substantial economic losses for farmers.Therefore,accurately identifying early symptoms of tomato early blight is essential for the scientific prevention and control of this disease.Additionally,visualizing affected areas can provide precise guidance for farmers,effectively reducing economic losses.This study combines hyperspectral imaging technology with machine learning algorithms to develop a model for the early identification of symptoms of tomato early blight,facilitating early detection of the disease and visual localization of affected areas.Methods To address noise interference present in hyperspectral images,robust principal component analysis(RPCA)is employed for effective denoising,enhancing the accuracy of subsequent analyses.To avoid insufficient information representation caused by the subjective selection of regions of interest,the Otsu’s thresholding method is utilized to extract tomato leaves effectively from the background,with the average spectrum of the entire leaf taken as the primary object of study.Furthermore,a comprehensive spectral preprocessing workflow is established by integrating multivariate scatter correction(MSC)and standardization methods,ensuring the reliability and effectiveness of the data.Based on the processed spectral data,a discriminant model utilizing a linear kernel function support vector machine(SVM)is constructed,focusing on characteristic wavelengths to improve the model"s discriminative capability.Results Compared to full-spectrum modeling,this approach results in an 8.33%increase in accuracy on the test set.After optimizing the parameters of the SVM model,when C=1.64,the accuracies of the training set and test set reach 91.67%and 94.44%,respectively,demonstrating a 1.19%increase in training set accuracy compared to the unoptimized model,while maintaining the same accuracy on the test set,effectively alleviating issues of underfitting.Conclusion This study successfully establishes an early discriminant model for tomato early blight using hyperspectral imaging and achieves visualization of early symptoms.Experimental results indicate that the SVM discriminant model based on characteristic wavelengths and a linear kernel function can effectively identify early symptoms of tomato early blight.Visualization of these symptoms in terms of disease probability allows for a more intuitive detection of early diseases and timely implementation of corresponding control measures.This visual analysis not only enhances the efficiency of disease identification but also provides farmers with more straightforward and practical information,aiding them in formulating more reasonable prevention strategies.These research findings provide valuable references for the early identification and visualization of plant diseases,holding significant practical implications for monitoring,identifying,and scientifically preventing crop diseases.Future research could further explore how to apply this model to disease detection in other crops and how to integrate IoT technology to create intelligent disease monitoring systems,enhancing the scientific and efficient management of crops.
作者
鲍浩
黄莉
张艳
庞浩
BAO Hao;HUANG Li;ZHANG Yan;PANG Hao(Engineering Research Centre for Non-Destructive Testing of Agricultural Products,College of Computer Science,Guiyang University,Guiyang 550005,China;School of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China)
出处
《生物化学与生物物理进展》
北大核心
2025年第2期513-524,共12页
Progress In Biochemistry and Biophysics
基金
国家自然科学基金(62265003,62141501)资助项目。
关键词
高光谱成像
机器学习
遗传算法
番茄早疫病
早期病状
可视化
hyperspectral imaging
machine learning
genetic algorithm
tomato early blight
early disease
visual analysis
