研究目的
To develop a method for spatially referencing time series of hyperspectral images to trace plant disease symptoms back in time, enabling the investigation of early, invisible states of diseases such as brown rust and septoria tritici blotch on wheat.
研究成果
The developed method successfully enables spatial referencing of hyperspectral images, allowing for the tracking of disease symptoms over time, including early invisible stages. It improves labeling and spectral analysis, providing new insights into disease dynamics for brown rust and septoria tritici blotch. Future work should focus on adapting the method for 3D scenarios and high-throughput applications without markers.
研究不足
The method requires the use of artificial markers on leaves, which may affect leaf tissue or disease development. It is limited to 2D leaf surfaces and may not handle 3D complexities or large distortions well. The spatial resolution of the camera restricts applicability to certain spectral regions, such as shortwave-infrared. The transformation models, while robust, can be less stable at leaf borders and may not fully represent local changes like wilting.
1：Experimental Design and Method Selection:
The study used a hyperspectral line scanning spectrograph (ImSpector V10E) to capture images in the VISNIR range (400-1000 nm). A non-linear 2D polynomial transformation model was selected for spatial referencing based on reference points (white spots on leaves) to handle leaf movements and growth. The algorithm involved background segmentation, reference point detection, matching using RanSaC, and spatial transformation.
2：Sample Selection and Data Sources:
Wheat plants (cv. Taifun and Chamsin) were grown in greenhouse conditions and inoculated with Zymoseptoria tritici and Puccinia triticina. Time series measurements were taken for brown rust (2-12 days after inoculation) and septoria tritici blotch (15-27 days after inoculation).
3：List of Experimental Equipment and Materials:
Hyperspectral camera (ImSpector V10E), ASD-Pro-Lamps for illumination, motorized line scanner, SpectralCube software, barium sulphate white reference, blue cardboard background, white color spots as reference points, strings for leaf fixation, and software tools (ENVI 4.6 + IDL 7.0, Matlab 2013a with Image Processing Toolbox).
4：6 + IDL 0, Matlab 2013a with Image Processing Toolbox).
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: Leaves were fixed horizontally in a tray with strings and reference points applied. Hyperspectral images were captured, normalized using white and dark references, and processed through the referencing algorithm steps (segmentation, point detection, matching, transformation). Vegetation Indices (NDVI, PRI, ARI) were calculated for spectral analysis.
5：Data Analysis Methods:
Root Mean Square Error (RMSE) was used to evaluate transformation accuracy. Random Forest algorithm was employed for classification and segmentation. Spectral indices were analyzed to track disease development over time.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
