A simple algorithm outperforms a machine learning approach for quantifying spittlebug damage in tropical grasses

In the extensive livestock systems of tropical America, host-plant resistance has proven to be the most efficient strategy for integrated pest management in forage grasses (i.e., Urochloa hybrids and Megathyrsus maximus) to spittlebug (Hemiptera: Cercopidae) attack. Precise and efficient quantificat...

Full description

Bibliographic Details
Main Authors: Ruiz-Hurtado, Andres Felipe, Espitia, Paula, Cardoso, Juan Andres, Jauregui, Rosa Noemi
Format: Poster
Language:Inglés
Published: 2024
Subjects:
Online Access:https://hdl.handle.net/10568/155544
_version_ 1855525869553975296
author Ruiz-Hurtado, Andres Felipe
Espitia, Paula
Cardoso, Juan Andres
Jauregui, Rosa Noemi
author_browse Cardoso, Juan Andres
Espitia, Paula
Jauregui, Rosa Noemi
Ruiz-Hurtado, Andres Felipe
author_facet Ruiz-Hurtado, Andres Felipe
Espitia, Paula
Cardoso, Juan Andres
Jauregui, Rosa Noemi
author_sort Ruiz-Hurtado, Andres Felipe
collection Repository of Agricultural Research Outputs (CGSpace)
description In the extensive livestock systems of tropical America, host-plant resistance has proven to be the most efficient strategy for integrated pest management in forage grasses (i.e., Urochloa hybrids and Megathyrsus maximus) to spittlebug (Hemiptera: Cercopidae) attack. Precise and efficient quantification of spittlebug damage is crucial for evaluation and selection of resistant and tolerant genotypes in the Urochloa spp. and M. maximus breeding programmes at CIAT. Traditional methods rely on visual inspection by experts, which is a time-consuming and resource-intensive process. Recent advancements in image processing offer the potential for automated high-throughput (HTP) analyses pipelines.  The proposed pipeline involves image pre-processing (normalisation, feature enhancement, and plant segmentation) followed by a damage segmentation algorithm. Considering the large data volumes in breeding trials, where five replicates of ~150 genotypes are assessed to spittlebug damage, often with limited availability of ground truth data, unsupervised learning approaches like clustering are preferred for damage segmentation. Furthermore, real-world image acquisition introduces challenges due to variability in lighting, noise, and lack of standardisation. The objective of this study was to compare a simple algorithm (Heckbert’s median-cut colour quantisation) with the k-means unsupervised machine learning approach for quantification of plant damage (green/chlorotic leaf tissue) by spittlebugs using colour images. Our results showed that Heckbert’s median-cut colour quantisation delivers similar results of quantification of plant damage to those obtained by K-means, yet, at a faster speed and less usage of CPU processing. We conclude that Heckbert's median-cut colour quantisation provides a computationally efficient and accurate solution for HTP spittlebug damage analysis of tropical forage grasses in CPU resource-constrained devices. This will facilitate the implementation of automated image analyses of spittlebug damage of tropical forage grasses for researchers working with old computers or mobile devices.
format Poster
id CGSpace155544
institution CGIAR Consortium
language Inglés
publishDate 2024
publishDateRange 2024
publishDateSort 2024
record_format dspace
spelling CGSpace1555442025-11-05T11:43:36Z A simple algorithm outperforms a machine learning approach for quantifying spittlebug damage in tropical grasses Ruiz-Hurtado, Andres Felipe Espitia, Paula Cardoso, Juan Andres Jauregui, Rosa Noemi phenotyping fenotipado algorithms damage image processing algoritmo procesamiento digital de imágenes daño a las plantas segmentation In the extensive livestock systems of tropical America, host-plant resistance has proven to be the most efficient strategy for integrated pest management in forage grasses (i.e., Urochloa hybrids and Megathyrsus maximus) to spittlebug (Hemiptera: Cercopidae) attack. Precise and efficient quantification of spittlebug damage is crucial for evaluation and selection of resistant and tolerant genotypes in the Urochloa spp. and M. maximus breeding programmes at CIAT. Traditional methods rely on visual inspection by experts, which is a time-consuming and resource-intensive process. Recent advancements in image processing offer the potential for automated high-throughput (HTP) analyses pipelines.  The proposed pipeline involves image pre-processing (normalisation, feature enhancement, and plant segmentation) followed by a damage segmentation algorithm. Considering the large data volumes in breeding trials, where five replicates of ~150 genotypes are assessed to spittlebug damage, often with limited availability of ground truth data, unsupervised learning approaches like clustering are preferred for damage segmentation. Furthermore, real-world image acquisition introduces challenges due to variability in lighting, noise, and lack of standardisation. The objective of this study was to compare a simple algorithm (Heckbert’s median-cut colour quantisation) with the k-means unsupervised machine learning approach for quantification of plant damage (green/chlorotic leaf tissue) by spittlebugs using colour images. Our results showed that Heckbert’s median-cut colour quantisation delivers similar results of quantification of plant damage to those obtained by K-means, yet, at a faster speed and less usage of CPU processing. We conclude that Heckbert's median-cut colour quantisation provides a computationally efficient and accurate solution for HTP spittlebug damage analysis of tropical forage grasses in CPU resource-constrained devices. This will facilitate the implementation of automated image analyses of spittlebug damage of tropical forage grasses for researchers working with old computers or mobile devices. 2024-09-12 2024-10-24T04:47:39Z 2024-10-24T04:47:39Z Poster https://hdl.handle.net/10568/155544 en Open Access application/pdf Ruiz-Hurtado, A.F.; Espitia, P.; Cardoso, J.A.; Jauregui, R.N. (2024) A simple algorithm outperforms a machine learning approach for quantifying spittlebug damage in tropical grasses. Poster prepared for Tropentag: Explore opportunities... for managing natural resources and a better life for all, on 11-13 September 2024 in Vienna (Austria). 1 p.
spellingShingle phenotyping
fenotipado
algorithms
damage
image processing
algoritmo
procesamiento digital de imágenes
daño a las plantas
segmentation
Ruiz-Hurtado, Andres Felipe
Espitia, Paula
Cardoso, Juan Andres
Jauregui, Rosa Noemi
A simple algorithm outperforms a machine learning approach for quantifying spittlebug damage in tropical grasses
title A simple algorithm outperforms a machine learning approach for quantifying spittlebug damage in tropical grasses
title_full A simple algorithm outperforms a machine learning approach for quantifying spittlebug damage in tropical grasses
title_fullStr A simple algorithm outperforms a machine learning approach for quantifying spittlebug damage in tropical grasses
title_full_unstemmed A simple algorithm outperforms a machine learning approach for quantifying spittlebug damage in tropical grasses
title_short A simple algorithm outperforms a machine learning approach for quantifying spittlebug damage in tropical grasses
title_sort simple algorithm outperforms a machine learning approach for quantifying spittlebug damage in tropical grasses
topic phenotyping
fenotipado
algorithms
damage
image processing
algoritmo
procesamiento digital de imágenes
daño a las plantas
segmentation
url https://hdl.handle.net/10568/155544
work_keys_str_mv AT ruizhurtadoandresfelipe asimplealgorithmoutperformsamachinelearningapproachforquantifyingspittlebugdamageintropicalgrasses
AT espitiapaula asimplealgorithmoutperformsamachinelearningapproachforquantifyingspittlebugdamageintropicalgrasses
AT cardosojuanandres asimplealgorithmoutperformsamachinelearningapproachforquantifyingspittlebugdamageintropicalgrasses
AT jaureguirosanoemi asimplealgorithmoutperformsamachinelearningapproachforquantifyingspittlebugdamageintropicalgrasses
AT ruizhurtadoandresfelipe simplealgorithmoutperformsamachinelearningapproachforquantifyingspittlebugdamageintropicalgrasses
AT espitiapaula simplealgorithmoutperformsamachinelearningapproachforquantifyingspittlebugdamageintropicalgrasses
AT cardosojuanandres simplealgorithmoutperformsamachinelearningapproachforquantifyingspittlebugdamageintropicalgrasses
AT jaureguirosanoemi simplealgorithmoutperformsamachinelearningapproachforquantifyingspittlebugdamageintropicalgrasses