Articles
Development of a high throughput phenotyping method to determine blueberry time-to-mature and yield using neural networks
Article number
1440_21
Pages
151 – 156
Language
English
Abstract
Time-to-mature and yield are important traits for blueberry breeding.
Proper determination of the time-to-mature of blueberry cultivars and breeding lines informs the harvest window which ensures that the fruits are harvested at optimum maturity and quality.
On the other hand, high yielding crops bring in high profit per acre of planting.
Previous data collection of these two traits were limited by categorical visual ratings for breeding programs lacking sufficient labor source.
In this study, we demonstrated an object detection model that can be used to estimate harvest window and yields in breeding lines overcoming the labor constraints of obtaining high frequency data.
Visual images of ten blueberry bushes were collected, annotated, and used to train a deep neural network object detection model (YOLOv8) to recognize mature and immature berries.
A linear regression was performed for the berry numbers predicted from the model and that of calculated from berry weight.
The results showed that the model detects approximately 20% of the berries on the bushes.
The Pearson correlation coefficient for mature and immature berries were 0.7 and 0.82, respectively, indicating 70 and 82% of variation in the data sets collected from mature and immature berries can be explained by the linear model.
The correlation coefficient of the ratio between mature and immature berries was 92%. The image model was further applied to 91 breeding lines.
The model was validated with a mAP50 of 0.68. Precision and recall for ripe berries were 0.70 and 0.80, respectively, and for green berries was 0.63 and 0.52, respectively.
This model will be useful for facilitating selecting breeding lines with high yield and desirable harvest window.
It can also be integrated into site-specific crop management system to facilitate growers to determine the optimum time for harvest.
Proper determination of the time-to-mature of blueberry cultivars and breeding lines informs the harvest window which ensures that the fruits are harvested at optimum maturity and quality.
On the other hand, high yielding crops bring in high profit per acre of planting.
Previous data collection of these two traits were limited by categorical visual ratings for breeding programs lacking sufficient labor source.
In this study, we demonstrated an object detection model that can be used to estimate harvest window and yields in breeding lines overcoming the labor constraints of obtaining high frequency data.
Visual images of ten blueberry bushes were collected, annotated, and used to train a deep neural network object detection model (YOLOv8) to recognize mature and immature berries.
A linear regression was performed for the berry numbers predicted from the model and that of calculated from berry weight.
The results showed that the model detects approximately 20% of the berries on the bushes.
The Pearson correlation coefficient for mature and immature berries were 0.7 and 0.82, respectively, indicating 70 and 82% of variation in the data sets collected from mature and immature berries can be explained by the linear model.
The correlation coefficient of the ratio between mature and immature berries was 92%. The image model was further applied to 91 breeding lines.
The model was validated with a mAP50 of 0.68. Precision and recall for ripe berries were 0.70 and 0.80, respectively, and for green berries was 0.63 and 0.52, respectively.
This model will be useful for facilitating selecting breeding lines with high yield and desirable harvest window.
It can also be integrated into site-specific crop management system to facilitate growers to determine the optimum time for harvest.
Publication
Authors
Y. Chu, J. Maleski, J. Zhang
Keywords
computer vision, YOLO8, fruit detection, harvest window, blueberry breeding
Online Articles (74)