Explainable deep learning in plant phenotyping.

Area of Science:

• Agricultural Science
• Computer Science
• Plant Biology

Background:

• Climate change and population growth threaten global food security.
• Plant phenotyping accelerates crop breeding and management but relies on complex deep learning models.
• Deep learning models in phenotyping are often 'black boxes,' lacking interpretability.

Purpose of the Study:

• To review the application of Explainable AI (XAI) in plant phenotyping.
• To highlight the benefits of XAI for understanding image-based phenotypic data.
• To encourage the integration of XAI in future plant science research.

Main Methods:

• Literature review of existing XAI studies in plant phenotyping and related fields.
• Analysis of how XAI can elucidate deep learning model representations.
• Synthesis of findings to guide plant researchers in adopting XAI.

Main Results:

• XAI offers a pathway to interpret deep learning models used in plant phenotyping.
• Interpretable models enhance the explanation of model decisions and feature relevance.
• XAI can improve the trustworthiness of image-based phenotypic data for agriculture.

Conclusions:

• XAI is crucial for advancing plant phenotyping by making deep learning models transparent.
• Integrating XAI will empower breeders and growers with reliable data for crop improvement.
• XAI fosters trust in AI-driven insights for sustainable food production.