Determination of the area index of lettuce leaves with a monocular camera

Abstract

This study aims to develop a pixel value analysis method using a monocular camera to determine different growth stages of lettuce plants. After the lettuce plants have been detected in the images obtained using the YOLOv4 (You Only Look Once Version 4) object detection algorithm, the leaf area index for each detected lettuce plant using the HSV (Hue, Saturation, Value) colour space has been calculated. The leaf area index serves as a fundamental metric in the analysis, aiding in accurately measuring the size of the lettuce plants. For the size estimation approach, a dataset containing HSV-calculated max area pixel index values of lettuce plants grown from 1 to 7 weeks has been used. By clustering pixel values using the Gaussian Mixture Models (GMM) algorithm, the cluster representing 1-week-old lettuce plants with the lowest pixel values has been identified, while the cluster representing 7-week-old lettuce plants had the highest pixel values. This process was repeated for each week, resulting in distinct clusters corresponding to specific weeks of lettuce growth. By associating the detected lettuce plants with their respective clusters, it was possible to infer the growth period and readiness for harvesting for each plant. This method offers valuable insights into monitoring lettuce growth and optimising harvesting schedules at different stages for lettuce farmers and agricultural researchers through non-intrusive imaging techniques. This study showcases the potential of computer vision and machine learning algorithms in transforming traditional agricultural practices into more efficient and data-driven processes. The conducted experiments demonstrate the successful integration of a monocular camera into a smart agriculture system for lettuce harvest detection. Through the combination of object detection using the YOLOv4 algorithm and area estimation using the HSV colour space and leaf area index, accurate and cost-effective size calculations have been achieved. The integration of Gaussian Mixture Model clustering with the dataset further enhanced the precision of the lettuce growth and harvest predictions.

Keywords

• artificial intelligence
• image processing
• hydroponics agriculture
• automation