摘要：AbstractPretreatment is an important step before analyzing THz spectral image. In this paper, maize seed was taken as an example to focus on exploring pretreatment methods and steps for THz optical information extracting of different seeds tissue structures (embryo and endosperm) based on THz time-domain spectral images. Firstly, the double Gaussian filter was used to de-noise THz spectral image of seeds. Then, the image reconstruction methods based on Time Slice, Diff (Differential) Time Slice and Peak-Peak in Slice were compared to enhance the visual representation of different tissues of seeds. Finally, the Sober edge detection algorithm was used to extract the embryo and endosperm regions of the reconstructed image, and calculated mean value of the THz time-domain pulse in different regions as representative information of the embryo and endosperm. The experimental results show that the double Gaussian filter denoising, Diff Time Slice image reconstruction and Sober edge detection algorithm can provide a feasible technical basis and reference for pretreatment methods for the content and distribution of seed chemical components based on THz spectral image.
关键词：KeywordsPrecision AgricultureCrop PhenotypingSensingAutomation and Robotics for Post-Harvest
