A labeled spectral dataset with cassava disease occurrences using virus titre determination protocol

Area of Science:

• Agricultural Science
• Plant Pathology
• Remote Sensing

Background:

• Traditional cassava disease diagnosis relies on visual inspection, often too late for effective intervention.
• Visible symptoms may not appear until irreversible damage has occurred to the edible root.
• Spectral information offers a potential method for detecting diseases before visual manifestation.

Purpose of the Study:

• To develop a novel dataset of spectral and image data for cassava disease detection.
• To investigate the potential of spectral data for early diagnosis of cassava diseases, including Cassava Brown Streak Disease and Cassava Mosaic Disease.
• To establish a protocol for collecting multimodal crop data in controlled and open-field environments.

Main Methods:

• Collected visible and near-infrared spectra and leaf imagery from healthy and diseased cassava plants.
• Acquired biochemical data as ground truth for disease assessment.
• Utilized expert disease scoring (1-5) for plant leaves over 19 weeks (screenhouse) and 15 weeks (open field).

Main Results:

• The dataset includes spectral and image data correlated with biochemical ground truth and expert disease scores.
• Data collection spanned extended periods to capture disease progression.
• The study hypothesizes that spectral data can identify diseased crops lacking visible symptoms.

Conclusions:

• The novel dataset supports research into early disease detection in cassava.
• Spectral analysis holds promise for non-invasive, early diagnosis of cassava diseases.
• Early detection via spectral data can facilitate timely intervention, potentially saving crops and improving yields.