Optimizing tomato plant phenotyping detection: Boosting YOLOv8 architecture to tackle data complexity

• DOI: 10.1016/j.compag.2024.108728
• Publication Date: 2024-02-09T18:10:58Z
• Authors: Firozeh Solimani; Angelo Cardellicchio; Giovanni Dimauro; Angelo Petrozza; Stephan Summerer; Francesco Cellini; Vito Renò
• Keywords: Boosting

Effective identification of tomato plant traits is crucial for timely monitoring and evaluating their growth harvest. However, conducting stress experiments on multiple genotypes introduces challenges due to the nature data. One these arises from an imbalanced sample distribution, potentially leading misclassification between classes disruptions in model recognition. This paper addresses effect by considering flowers, fruits, nodes proposing improved detection approach through data balancing. A novel data-balancing introduced this study overcome issue The proposed solution involves implementation a YOLOv8 deep learning model, which effectively detects plants. significantly enhances ability algorithm detect objects varying sizes within complex environments. To further bolster recognition capability targeted classes, integrates Squeeze-and-Excitation (SE) block attention module into its head architecture. strengthens giving increased studied thereby enhancing overall performance. results demonstrate that balancing successfully improves performance response challenges. When applying technique pre-training optimal weights obtained balanced data, SE-block showed significant improvements outcomes.