DM_CorrMatch: a semi-supervised semantic segmentation framework for rapeseed flower coverage estimation using UAV...

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Rapeseed (Brassica napus L.) inflorescence coverage is a key metric for crop growth assessment and yield prediction.
• Unmanned Aerial Vehicle (UAV) imagery combined with semantic segmentation is standard for crop cover assessment.
• Irregular rapeseed inflorescence morphology poses significant segmentation challenges, especially with limited data.

Purpose of the Study:

• To develop a cost-effective, high-throughput method for accurate rapeseed inflorescence segmentation.
• To address challenges in segmentation accuracy under limited data conditions.
• To improve crop monitoring and aid in developing high-yield rapeseed cultivars.

Main Methods:

• Proposed a semi-supervised learning framework (DM_CorrMatch) utilizing strong and weak data augmentation.
• Leveraged the Denoising Diffusion Probabilistic Model (DDPM) for synthetic data generation in data-scarce scenarios.
• Introduced a novel Mamba-Deeplabv3+ network architecture for effective global and local feature extraction, handling complex backgrounds and varied poses.
• Implemented an automatic labeled data update strategy to mitigate erroneous labels.

Main Results:

• The DM_CorrMatch method achieved superior performance on the Rapeseed Flower Segmentation Dataset (RFSD).
• Achieved high segmentation accuracy with an Intersection over Union (IoU) of 0.886, Precision of 0.942, and Recall of 0.940.
• Outperformed four traditional and eleven deep learning segmentation methods.

Conclusions:

• The proposed semi-supervised learning approach with the Mamba-Deeplabv3+ architecture provides a robust solution for rapeseed inflorescence segmentation.
• The method effectively handles complex backgrounds and diverse inflorescence poses, offering a reliable tool for flower cover estimation.
• This technology can significantly enhance crop monitoring via UAVs and support the development of improved rapeseed varieties.