A specific fine-grained identification model for plasma-treated rice growth using multiscale shortcut convolutional...

Area of Science:

• Agricultural Technology
• Computer Vision
• Plant Science

Background:

• Low-temperature plasma technology offers an eco-friendly approach to enhance crop yield and quality.
• Distinguishing plasma-treated rice growth is crucial but under-researched.
• Traditional Convolutional Neural Networks (CNNs) have limitations in fine-grained identification tasks.

Purpose of the Study:

• To develop an efficient model for identifying plasma-treated rice growth.
• To leverage spatial and local information for improved fine-grain identification.
• To evaluate the performance of a novel Multiscale Shortcut CNN (MSCNN) model.

Main Methods:

• Collected 5000 images of rice at the tillering stage, including plasma-treated and control samples.
• Proposed an efficient Multiscale Shortcut CNN (MSCNN) model.
• Utilized cross-layer features and shortcuts for enhanced information extraction.

Main Results:

• The MSCNN model achieved high performance metrics: 92.64% accuracy, 90.87% recall, 92.88% precision, and 92.69% F1 score.
• MSCNN outperformed mainstream models in identification tasks.
• Ablation experiments confirmed that shortcuts significantly enhance precision.

Conclusions:

• The proposed MSCNN model is effective for identifying plasma-treated rice growth.
• The integration of shortcuts and multiscale features is key to the model's superior performance.
• This research contributes to the advancement of agricultural technology through AI-driven plant identification.