Sunpheno: A Deep Neural Network for Phenological Classification of Sunflower Images

Area of Science:

• Agricultural Science
• Computer Science
• Plant Biology

Background:

• Leaf senescence is critical for grain filling and yield, requiring synchronization with plant phenological stages.
• Accurate monitoring of phenological stages is essential for optimizing crop yields.

Purpose of the Study:

• To evaluate deep machine learning methods for assessing sunflower phenological stages using field images.
• To develop an accurate and efficient model for tracking leaf senescence and its impact on yield.

Main Methods:

• Five deep machine learning models were tested using cell phone images of sunflowers.
• The pre-trained resnet50 network was identified as the top-performing model.
• A database of 5000 expert-classified sunflower images was created.

Main Results:

• The resnet50-based model (Sunpheno) demonstrated superior accuracy and speed in evaluating phenological stages.
• Distinct phenological differences were observed between sunflower lines B481_6 and R453 during senescence.
• The generated model reduces subjectivity in field assessments of leaf senescence.

Conclusions:

• Deep learning, particularly the resnet50 model, offers a robust solution for objective phenological stage evaluation in sunflowers.
• The Sunpheno model can aid in correlating senescence parameters with crop performance and yield.
• Standardized image-based phenotyping can improve decision-making in agricultural research and practice.