Automatic estimation of heading date of paddy rice using deep learning

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Breeding

Background:

• Accurate estimation of paddy rice heading date is vital for crop adaptability assessment and yield determination in research.
• Manual visual crop examination is time-consuming and labor-intensive, necessitating efficient automated methods.

Purpose of the Study:

• To develop a quick and precise method for estimating the heading date of paddy rice.
• To automate the process of heading date detection, replacing laborious manual observation.

Main Methods:

• A pipeline was developed to detect flowering panicle regions from ground-level RGB images of paddy rice.
• Image classification using Convolutional Neural Networks (CNNs) was employed to analyze image data.
• Flowering panicle region counts were used as a proxy to estimate the crop's heading date.

Main Results:

• The proposed method achieved an estimation of heading date with a mean absolute error of less than 1 day.
• Performance was evaluated on time-series image sequences of three different rice varieties.
• The algorithm demonstrated improved accuracy and general versatility compared to previous methods.

Conclusions:

• An efficient heading date estimation method for rice crops using time-series RGB images under natural field conditions was established.
• The developed method can reliably replace manual observation for detecting rice heading dates.
• This approach offers a significant advancement in precision agriculture for rice cultivation.