Digital Assessment of Banana Weevil (Cosmopolites sordidus Germar) Resistance as Compared with Expert Visual Assessment

Abstract

Accurately assessing weevil damage is critical when evaluating banana germplasm to identify genotypes resistant to the banana weevil (Cosmopolites sordidus), for use as elite parents in the banana breeding pipeline or evaluating breeding products. Visual observation remains the most common phenotyping approach but limited by individual bias. This study investigated the potential of image analyses as precise and objective alternatives for assessing weevil damage on the banana corm. Phenotyping trials were set up as partially replicated (P-rep) designs with 22 tissue culture-generated genotypes raised in pots and infested with banana weevils. At termination, the percentage score of weevil damage to the corms was evaluated by visual observation and image analysis using ImageJ and machine learning. In total, 370 high-quality images were assessed for weevil damage using ImageJ and machine learning. On average, damage scores from visual observation were 5.52% and 3.88% higher than ImageJ and machine learning respectively. There was a proportional trend with visual observation agreeing closely to image analyses for smaller scores and but less for larger scores. The results show that both ImageJ and machine learning exhibited a strong level of agreement and are interchangeable with consistent, reliable, and repeatable measurements. In conclusion, to avoid individual bias and subjectivity arising from visual observation, we recommend the use of either ImageJ or machine learning when scoring weevil damage in the banana corm.

Plant phenotyping relevance

バナナ果茎のゾウムシ被害という植物状態を、画像解析と機械学習で定量化し、目視法との一致度・再現性を検証した方法研究である。

Code and data availability

The paper describes 370 banana corm images, an ilastik classifier, and a YOLOv8 model, but no public deposit of the images, annotations, trained model, or analysis code is stated. Supplementary materials are only available via the preprint's own page (not an allowed URL), and the only public URLs cited (Ultralytics, DO

No evidence-backed public reproduction asset is currently recorded.