SSD-YOLO: a lightweight network for rice leaf disease detection.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Rice leaf diseases critically affect crop yield and quality.
• Traditional diagnostic methods are subjective and error-prone.
• Accurate and efficient disease detection is vital for rice cultivation.

Purpose of the Study:

• To develop an improved YOLOv8-based method for accurate rice disease detection.
• To enhance feature extraction and sampling accuracy for complex disease patterns.
• To boost model performance in challenging environmental conditions.

Main Methods:

• Implemented a novel SSD-YOLO model integrating Squeeze-and-Excitation Network (SENet) attention.
• Utilized a Dynamic Sample (DySample) module for improved upsampling accuracy.
• Employed Shape-aware Intersection over Union (ShapeIoU) Loss for enhanced detection.
• Trained and validated the model on a dataset of 3000 rice leaf disease images.

Main Results:

• SSD-YOLO achieved high detection accuracies: 87.52% for brown spot, 99.48% for blast, and 98.99% for bacterial blight.
• Significant improvements over original YOLOv8 were observed: 11.11% for brown spot, 1.73% for blast, and 3.81% for bacterial blight.
• The model is compact (6MB) while demonstrating enhanced detection accuracy and speed.

Conclusions:

• The SSD-YOLO model offers a robust and efficient solution for timely rice disease identification.
• The integration of SENet, DySample, and ShapeIoU loss significantly boosts detection performance.
• This approach provides valuable support for precision agriculture and disease management in rice farming.