A CNN-based model to count the leaves of rosette plants (LC-Net)

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Plant image analysis is crucial for plant phenotyping, aiding in growth assessment and forecasting.
• Leaf area segmentation and counting are key metrics for measuring plant growth.
• Existing methods require robust models for accurate leaf counting.

Purpose of the Study:

• To develop a novel convolutional neural network (CNN)-based leaf counting model, termed LC-Net.
• To enhance leaf counting accuracy by incorporating segmented leaf parts as additional input.
• To evaluate LC-Net's performance against other state-of-the-art CNN models.

Main Methods:

• Developed LC-Net, a CNN-based model for leaf counting.
• Utilized SegNet for segmenting leaf parts, outperforming DeepLab V3+, Fast FCN, U-Net, and Refine Net.
• Trained and tested LC-Net on combined Computer Vision Problems in Plant Phenotyping (CVPPP) and KOMATSUNA datasets.

Main Results:

• LC-Net demonstrated superior performance in leaf counting compared to other tested CNN models.
• Both subjective and numerical evaluations confirmed LC-Net's effectiveness.
• The inclusion of segmented leaf parts significantly improved the model's accuracy.

Conclusions:

• LC-Net represents a significant advancement in CNN-based leaf counting for plant phenotyping.
• The model's ability to leverage segmented leaf data offers a more precise method for plant growth analysis.
• LC-Net shows great potential for applications in precision agriculture and botanical research.