Hyperspectral imaging for detection of russet mite infestation and drought stress in the greenhouse tomato cultivation

The importance of automated stress monitoring is growing, as early detection of infestation events can significantly reduce yield losses and the use of chemical synthetic plant protection products. This approach aligns with the agricultural policies of the European Union and Germany.

Hyperspectral Imaging (HSI) is a key non-destructive technique for detecting plant stress and identifying symptoms of pest infestations and abiotic stress. While demand for HSI applications is increasing, its practical implementation remains challenging. Most existing studies have been conducted under controlled laboratory conditions, limiting direct transferability to real-world environments such as greenhouses.

In greenhouse settings, uncontrolled factors—for instance variable ambient light and plant self-shadowing—pose significant challenges to accurate spectral measurements. To address these issues, this study utilized a VNIR hyperspectral camera (500–1000 nm) with an integrated VNIR broadband LED illumination system to ensure consistent lighting conditions. We collected spectral data from tomato plants affected by russet mite infestation and drought stress under real greenhouse conditions.

The data were processed using key vegetation indices which were calculated and analysed to identify spectral signatures associated with stress symptoms. Additionally, machine learning algorithms were applied to develop predictive models for early stress detection.

In our project, we are performing this approach to include a broader range of plant stresses—including different pests, pathogens, and abiotic stressors—to enhance the robustness and generalizability of the detection system. The ultimate goal is to improve precision crop management in greenhouses through early, automated, and non-destructive stress monitoring.

In the poster, the current results from the russet mite and drought stress trials will be presented. The poster shows the effectiveness of vegetation indices, spectral responses, and model performance, in distinguishing stress types based on the two aforementioned stressors.