Seedling maize counting method in complex backgrounds based on YOLOV5 and Kalman filter tracking algorithm.

Area of Science:

• Agricultural Science
• Computer Vision
• Remote Sensing

Background:

• Maize population density significantly impacts crop yield and quality.
• Manual counting is inefficient, and traditional image processing struggles with complex field backgrounds.
• Accurate and timely estimation of maize plant density is essential for agricultural management.

Purpose of the Study:

• To develop a deep-learning-based method for accurate maize plant counting.
• To overcome the limitations of manual counting and traditional image processing techniques.
• To provide a robust solution for estimating maize population density in complex environments.

Main Methods:

• Collected maize field image datasets using a low-altitude UAV.
• Trained a real-time maize plant detection model using the YOLOV5 object detection architecture.
• Implemented maize plant tracking and counting via Hungarian matching and Kalman filtering algorithms.

Main Results:

• The SE-YOLOV5m detection model achieved an average precision (mAP@0.5) of 90.66% on the test dataset.
• Maize plant counts from the model showed a high correlation (R = 0.92) with manual counts across multiple locations.
• The method demonstrated effectiveness in complex background environments.

Conclusions:

• The proposed deep-learning method accurately identifies and counts maize plants, even in challenging field conditions.
• This approach offers a reliable and efficient alternative to manual counting for determining maize population density.
• The study provides a foundation for rapid acquisition of maize plant population data, supporting precision agriculture.