Winter wheat ear counting based on improved YOLOv7x and Kalman filter tracking algorithm with video streaming

Area of Science:

• Agricultural Engineering
• Computer Vision
• Machine Learning

Background:

• Accurate wheat ear counting is crucial for yield prediction, breeding, and management.
• Existing methods struggle with real-time, high-resolution Unmanned Aerial Vehicle (UAV) video data.

Purpose of the Study:

• To develop an accurate and real-time wheat ear detection and counting model for UAV video.
• To improve upon existing deep learning models for agricultural applications.

Main Methods:

• Integrated a Space to Depth (SPD) module into the YOLOv7x deep learning model, creating YOLOv7xSPD.
• Utilized Normalized Gaussian Wasserstein Distance (NWD) Loss function for enhanced detection.
• Combined YOLOv7xSPD with Kalman filter tracking and Hungarian matching for video-based counting (YOLOv7xSPD Counter).

Main Results:

• YOLOv7xSPD achieved high precision (95.85%), recall (94.71%), F1 score (95.28%), and AP (94.99%).
• The model demonstrated superior performance compared to YOLOv7x and other object detection models (Faster RCNN, SSD, YOLOv5s, YOLOv7).
• YOLOv7xSPD Counter processed high-resolution video (3840x2160) at approximately 5.5 FPS with high correlation (R² = 0.99) to ground truth.

Conclusions:

• The YOLOv7xSPD Counter model offers a robust solution for real-time wheat ear counting from UAV video.
• This technology provides a valuable tool for wheat yield prediction, genetic breeding, and optimized planting management.