Reliable plant segmentation under variable greenhouse illumination conditions

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F61989592%3A15640%2F25%3A73627759" target="_blank" >RIV/61989592:15640/25:73627759 - isvavai.cz</a>

Výsledek na webu

<a href="https://www.sciencedirect.com/science/article/pii/S0168169924011025?via%3Dihub" target="_blank" >https://www.sciencedirect.com/science/article/pii/S0168169924011025?via%3Dihub</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1016/j.compag.2024.109711" target="_blank" >10.1016/j.compag.2024.109711</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Reliable plant segmentation under variable greenhouse illumination conditions

Popis výsledku v původním jazyce

The effectiveness of image-based sensors in capturing meaningful morphological and physiological information relies on the accuracy of the segmentation strategy used. Fluctuating light conditions and background noise can reduce the reliability of plant tissue pixel classification using color-based threshold strategies even under semicontrolled greenhouse conditions. The implementation of multiple criteria classification strategies can provide a more reliable segmentation of focal objects regardless of the environmental influence, but some approaches can be time-consuming and computationally intensive. To identify the most accurate method for semantic segmentation of lettuce plants grown in a greenhouse under varying light conditions, we compared a single-color threshold baseline strategy using Otsu segmentation with two advanced machine learning segmentation techniques: Random Forest classification and U-Net Convolutional Neural Network (CNN). Each method was trained and evaluated to determine its effectiveness in accurately segmenting lettuce plants based on their performance under different light conditions. The U-Net model demonstrated superior performance on images captured under varying illumination conditions and crop developmental stages. U-Net achieved average Intersection over Union (IoU) and F1 Score values of 0.940 and 0.969, respectively. In contrast, the Random Forest yielded average IoU and F1 Score values of 0.842 and 0.913, while Otsu segmentation produced values of 0.842 and 0.906. These results highlight the efficacy of the U-Net model in segmentation tasks under variable greenhouse conditions. Additionally, the U-Net model automatic feature extraction step outperformed the inefficient feature extraction step involved in training a Random Forest-based segmentation. Despite a relatively small set of training images, U-Net&apos;s architecture more effectively resolved the segmentation inaccuracies associated with varying greenhouse light conditions, demonstrating U-Net&apos;s advantage for non-robust, color-based segmentation.

Název v anglickém jazyce

Reliable plant segmentation under variable greenhouse illumination conditions

Popis výsledku anglicky

The effectiveness of image-based sensors in capturing meaningful morphological and physiological information relies on the accuracy of the segmentation strategy used. Fluctuating light conditions and background noise can reduce the reliability of plant tissue pixel classification using color-based threshold strategies even under semicontrolled greenhouse conditions. The implementation of multiple criteria classification strategies can provide a more reliable segmentation of focal objects regardless of the environmental influence, but some approaches can be time-consuming and computationally intensive. To identify the most accurate method for semantic segmentation of lettuce plants grown in a greenhouse under varying light conditions, we compared a single-color threshold baseline strategy using Otsu segmentation with two advanced machine learning segmentation techniques: Random Forest classification and U-Net Convolutional Neural Network (CNN). Each method was trained and evaluated to determine its effectiveness in accurately segmenting lettuce plants based on their performance under different light conditions. The U-Net model demonstrated superior performance on images captured under varying illumination conditions and crop developmental stages. U-Net achieved average Intersection over Union (IoU) and F1 Score values of 0.940 and 0.969, respectively. In contrast, the Random Forest yielded average IoU and F1 Score values of 0.842 and 0.913, while Otsu segmentation produced values of 0.842 and 0.906. These results highlight the efficacy of the U-Net model in segmentation tasks under variable greenhouse conditions. Additionally, the U-Net model automatic feature extraction step outperformed the inefficient feature extraction step involved in training a Random Forest-based segmentation. Despite a relatively small set of training images, U-Net&apos;s architecture more effectively resolved the segmentation inaccuracies associated with varying greenhouse light conditions, demonstrating U-Net&apos;s advantage for non-robust, color-based segmentation.

Druh

J_imp - Článek v periodiku v databázi Web of Science

CEP obor

—

OECD FORD obor

40106 - Agronomy, plant breeding and plant protection; (Agricultural biotechnology to be 4.4)

Projekt

—

Návaznosti

I - Institucionalni podpora na dlouhodoby koncepcni rozvoj vyzkumne organizace

Rok uplatnění

2025

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název periodika

COMPUTERS AND ELECTRONICS IN AGRICULTURE

ISSN

0168-1699

e-ISSN

1872-7107

Svazek periodika

229

Číslo periodika v rámci svazku

February

Stát vydavatele periodika

GB - Spojené království Velké Británie a Severního Irska

Počet stran výsledku

13

Strana od-do

nestránkováno

Kód UT WoS článku

001388521000001

EID výsledku v databázi Scopus

2-s2.0-85211626946