Work

Automated hyperspectral data collection system for monitoring greenhouse plant health

Year: 2026

Type: article

Abstract: Purpose Hyperspectral sensing (HS) offers strong potential for plant health monitoring, but its application in greenhouse environments is limited by dynamic conditions and the lack of automated canopy-scale data collection. This study aimed to develop an automated HS sensing platform and preprocessing workflow to improve data quality under greenhouse conditions. Methods A mobile canopy-scale HS sensing system was developed by integrating boom-based positioning, depth-guided sensing distance control, sunlight monitoring, and automated calibration. The system was evaluated in a production-style greenhouse through five independent tomato bacterial spot experiments. Environmental effects on spectral measurements were examined, and spectra affected by transient lighting changes were removed. Preprocessing effectiveness was evaluated using sample-to-centroid distance analysis, spectral variability analysis, threshold-based classification metrics, and receiver operating characteristic-area under the curve (ROC-AUC) from logistic regression and linear SVM models. Results Seasonal greenhouse variability produced measurable spectral differences across experiments, while no meaningful AM and PM differences were observed between the two daily measurement times. Increased air moisture introduced higher near-infrared spectral noise, and short-term transportation affected early spectral responses. After preprocessing, healthy and inoculated groups showed reduced spectral variability. Despite the smaller quality-screened dataset, preprocessing improved all reported threshold-based classification metrics for both classifiers. ROC-AUC also increased, indicating improved threshold-independent discrimination between healthy and inoculated samples. Conclusion The automated HS sensing system improved greenhouse-collected spectral data quality and provided higher-quality data for downstream analysis. Although the linear baseline classifiers were not suitable for practical commercial disease diagnosis, systematic preprocessing improved the usability of canopy-scale HS data collected under dynamic greenhouse conditions.

Source: Smart Agricultural Technology

DOI: 10.1016/j.atech.2026.102331

Authors: Gaoshoutong Si, Peter Ling, Anna Louise Testen, Heping Zhu, Hongyoung Jeon

Institutions: The Ohio State University, Agricultural Research Service

Language: English

Topic: Smart Agriculture and AI

Subfield: Plant Science

Field: Agricultural and Biological Sciences

Domain: Life Sciences

SDG: Zero hunger

Open Access status: gold

Estimated APC paid: $1,770