基于高光谱成像和SSFNet的棉花黄萎病严重程度检测  

Cotton Verticillium Wilt Severity Detection Based on Hyperspectral Imaging and SSFNet

作　　者：吴念倚 仓浩 高秀文 李泳泉 谭菲 狄若愚 阮世伟 高攀[1] 吕新[2] WU Nian-yi;CANG Hao;GAO Xiu-wen;LI Yong-quan;TAN Fei;DI Ruo-yu;RUAN Shi-wei;GAO Pan;LÜXin(College of Information Science and Technology,Shihezi University,Shihezi 832003,China;College of Agriculture,Shihezi University,Shihezi 832003,China)

机构地区：[1]石河子大学信息科学与技术学院,新疆石河子832003 [2]石河子大学农学院,新疆石河子832003

出　　处：《光谱学与光谱分析》2025年第5期1300-1309,共10页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(62265015,61965014);八师石河子市重点领域创新团队计划(2023TD01)资助。

摘　　要：黄萎病对棉花产量和质量构成严重威胁。快速、准确地检测黄萎病严重程度是棉花黄萎病(CVW)控制的基础。现有的CVW检测主要集中在图像或光谱级别,忽略了特征融合的重要性,这限制了模型的性能。为解决此问题,提出了一种CVW等级检测方法图谱特征融合模型(SSFNet)。首先,选择对受感染植物像素变化敏感的LAB颜色空间进行数据增强,以丰富RGB图像的特征表示,并使用改进的ResNet网络构建图像特征提取模块。此外,还构建了光谱特征提取模块,并比较最小绝对收缩和选择算子(LASSO)以及主成分分析(PCA)算法选取的特征波段与全光谱之间的差异。最后,基于图像和光谱级别的探索构建了特征融合模型SSFNet。实验结果显示,相较于用单一数据类型特征进行检测,图谱特征融合方法SSFNet效果最好,F1分数达到了95.96%,展现了图谱特征融合方法与深度学习技术结合在CVW等级检测中的潜力。Verticillium wilt poses a severe threat to cotton yield and quality.Rapid and accurate detection of Verticillium wilt is essential for controlling cotton Verticillium wilt(CVW).Existing CVW detection methods mainly focus on the image or spectral level,overlooking the importance of feature fusion,which limits model performance.We propose a CVW grade detection method,spatial-spectral Fusion Network(SSFNet),to address this.First,we enhance the LAB color space,which is sensitive to pixel changes in infected plants,to enrich the feature representation of RGB images and use an improved ResNet network to build an image feature extraction module.Next,we construct a spectral feature extraction module based on the improved ResNet network and compare the performance of two common feature extraction methods:Least Absolute Shrinkage and Selection Operator(LASSO)and Principal Component Analysis(PCA).Finally,we build the feature fusion model SSFNet based on image and spectral level exploration.Experimental results show that SSFNet performs best compared to single data type features,with an F1 score of 95.96%,demonstrating the potential of image-spectral feature fusion methods combined with deep learning for CVW grade detection.

关 键 词：棉花黄萎病 图谱特征融合 高光谱成像 深度学习 

分 类 号：O657.3[理学—分析化学]