Identification of Soybean Mutant Lines Based on Dual-Branch CNN Model Fusion Framework Utilizing Images from...

Area of Science:

• Agricultural Science
• Computer Science
• Genetics

Background:

• Accurate identification of soybean mutant lines is vital for mutation breeding programs.
• Existing methods often focus on variety classification, facing challenges in distinguishing genetically similar mutant lines based on seeds alone.

Purpose of the Study:

• To develop an effective method for classifying soybean mutant lines using a dual-branch convolutional neural network (CNN).
• To fuse image features from both soybean pods and seeds for improved classification accuracy.

Main Methods:

• Designed a dual-branch CNN architecture integrating features from two identical single CNNs.
• Utilized four single CNNs (AlexNet, GoogLeNet, ResNet18, ResNet50) for feature extraction.
• Employed clustering and t-distributed stochastic neighbor embedding (t-SNE) for identifying similar lines and genetic relationships.

Main Results:

• Dual-branch CNNs significantly outperformed single CNNs in classification.
• The dual-ResNet50 fusion framework achieved a high classification rate of 90.22 ± 0.19%.
• Successfully identified closely related soybean mutant lines and their genetic connections.

Conclusions:

• The proposed dual-branch CNN approach offers a robust solution for soybean mutant line classification.
• Combining features from multiple plant organs (pods and seeds) enhances identification accuracy.
• This study provides a novel strategy for selecting potential lines in soybean mutation breeding and advances recognition technology.