HSI-PP: A flexible open-source software for hyperspectral imaging-based plant phenotyping

Hyperspectral imaging has become one of the most popular techniques for high-throughput plant phenotyping. Extracting and analyzing useful plant phenotypic traits from hyperspectral images represents a major bottleneck for plant science and breeding communities. This study aims to present a stand-alone easy-to-use software platform called HSI-PP to process and analyze hyperspectral images for high-throughput plant phenotyping. The HSI-PP software integrates pre-processing, feature extraction, and modeling functions. The application of HSI-PP is exemplified by investigating the response of different Arabidopsis thaliana genotypes to drought stress, and the impact of various imaging angles on predicting the canopy nitrogen content (CNC) of oilseed rape (Brassica napus L.). The results showed that HSI-PP can process 10 GB on an ordinary PC in time ranging from 30 to 73 min according to image size and the complexity of the pipeline. HSI-PP extracted multiple phenotyping traits (spectral, textural, and morphological) of Arabidopsis thaliana from a large image dataset (104 GB) within five hours. The fusion of these features achieved higher accuracy (94%) than only using spectral information (85%) as early as day 4 after drought stress treatment. For oilseed rape, about 384 GB image data was processed within eighteen hours, and it was found that the tilted imaging angle of 75 had the optimized PLSR fitting (0.83) to the ground truth. The results demonstrate that HSI-PP is a stand-alone, automated, and open-source hyperspectral image processing platform adapted to various applications in plant phenotyping without requiring professional programming skills to serve the plant research community.

Automated summary

Hyperspectral imaging is a popular technique for plant phenotyping, but extracting useful traits from the images is challenging. This study introduces HSI-PP, a standalone software platform that can process and analyze hyperspectral images for high-throughput plant phenotyping. The results demonstrate its efficiency and accuracy in extracting phenotypic traits from large image datasets.