Corn Seed Defect Detection Based on Watershed Algorithm and Two-Pathway Convolutional Neural Networks.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Automated quality control is crucial for agricultural products like corn seeds.
• Traditional methods for seed defect detection can be time-consuming and less accurate.
• Developing efficient, high-throughput methods for corn seed inspection is essential.

Purpose of the Study:

• To develop and evaluate a novel defect detection method for corn seeds.
• To leverage a two-pathway convolutional neural network (CNN) model combined with a watershed algorithm.
• To utilize both RGB and near-infrared (NIR) imaging for enhanced detection capabilities.

Main Methods:

• Acquisition of multispectral images (RGB and NIR) of corn seeds.
• Implementation of a watershed algorithm integrated with a two-pathway CNN model.
• Training and validation of the model for classifying defective versus defect-free seeds.

Main Results:

• The proposed method achieved high performance metrics: 95.63% average accuracy, 95.29% average recall, and 95.46% F1 score.
• The two-pathway CNN model demonstrated superiority over traditional one-pathway CNN methods using only RGB images.
• Analysis of parameter settings' influence on model training was conducted.

Conclusions:

• The developed object detection method is highly effective for corn seed defect detection.
• This approach offers a promising tool for high-throughput quality control in the corn seed industry.
• The integration of multispectral imaging and advanced CNN architectures enhances detection accuracy.