Efficient Wheat Head Segmentation with Minimal Annotation: A Generative Approach.

Area of Science:

• Computer Vision
• Machine Learning
• Agricultural Technology

Background:

• Supervised deep learning models require large annotated datasets, which are costly and time-consuming to create.
• Synthetic data can be used but suffers from a domain gap, leading to poor performance on real-world data.
• This domain gap hinders the practical application of deep learning in image processing tasks.

Purpose of the Study:

• To address the challenge of limited annotated data for deep learning model development.
• To bridge the domain gap between synthetic and real-world images using generative adversarial networks (GANs).
• To create a realistic annotated synthetic dataset for wheat head segmentation.

Main Methods:

• Computationally simulated a large-scale annotated dataset.
• Employed a generative adversarial network (GAN) to minimize the domain gap between simulated and real images.
• Utilized the enhanced synthetic dataset to train a deep-learning model for semantic segmentation.

Main Results:

• Developed a realistic annotated synthetic dataset for wheat head segmentation.
• Achieved a Dice score of 83.4% on an internal dataset.
• Obtained Dice scores of 79.6% and 83.6% on external datasets from the Global Wheat Head Detection datasets.

Conclusions:

• The proposed GAN-based approach effectively bridges the domain gap, enabling the use of synthetic data for training deep learning models.
• The method is highly effective for wheat head segmentation and shows potential for generalization to other crop types and dense imagery.
• This research facilitates the development of robust deep learning models even with limited real-world annotated data.