High-throughput spike detection in greenhouse cultivated grain
crops with attention mechanisms based deep learning models

J 2024

High-throughput spike detection in greenhouse cultivated grain crops with attention mechanisms based deep learning models

ULLAH, Sajid, Klára PANZAROVÁ, Martin TRTÍLEK, Matej LEXA, Vojtěch MÁČALA et. al.

Základní údaje

Originální název

High-throughput spike detection in greenhouse cultivated grain crops with attention mechanisms based deep learning models

Autoři

ULLAH, Sajid (586 Pákistán, domácí), Klára PANZAROVÁ, Martin TRTÍLEK, Matej LEXA (703 Slovensko, domácí), Vojtěch MÁČALA (203 Česká republika, domácí), Thomas ALTMANN, Kerstin NEUMANN, Jan HEJÁTKO (203 Česká republika, domácí), Markéta PERNISOVÁ (203 Česká republika, domácí) a Evgeny GLADILIN

Vydání

Plant Phenomics, AAAS, 2024, 2643-6515

Další údaje

Jazyk

angličtina

Typ výsledku

Článek v odborném periodiku

Obor

10201 Computer sciences, information science, bioinformatics

Stát vydavatele

Spojené státy

Utajení

není předmětem státního či obchodního tajemství

Odkazy

URL

Impakt faktor

Impact factor: 6.500 v roce 2022

Organizační jednotka

Přírodovědecká fakulta

DOI

http://dx.doi.org/10.34133/plantphenomics.0155

UT WoS

001231155500001

Klíčová slova anglicky

spike detection; high-throughput image analysis; Attention networks; Deep Neural Networks

Příznaky

Mezinárodní význam, Recenzováno

Změněno: 13. 9. 2024 15:55, Mgr. Markéta Pernisová, Ph.D.

Anotace

V originále

Detection of spikes is the first important step towards image-based quantitative assessment of crop yield. However, spikes of grain plants occupy only a tiny fraction of the image area and often emerge in the middle of the mass of plant leaves that exhibit similar colors as spike regions. Consequently, accurate detection of grain spikes renders, in general, a non-trivial task even for advanced, state-of-the-art deep learning neural networks (DNNs). To improve pattern detection in spikes, we propose architectural changes to Faster-RCNN (FRCNN) by reducing feature extraction layers and introducing a global attention module. The performance of our extended FRCNN-A vs. conventional FRCNN was compared on images of different European wheat cultivars, including ’difficult’ bushy phenotypes from two different phenotyping facilities and optical setups. Our experimental results show that introduced architectural adaptations in FRCNN-A helped to improve spike detection accuracy in inner regions. The mAP of FRCNN and FRCNN-A on inner spikes is 76.0% and 81.0%, respectively, while on the state-of-the-art detection DNNs, Swin Transformer mAP is 83.0%. As a lightweight network, FRCNN-A is faster than FRCNN and Swin Transformer on both baseline and augmented training datasets. On the FastGAN augmented dataset, FRCNN achieved mAP of 84.24%, FRCNN-A 85.0%, and the Swin Transformer 89.45%. The increase in mAP of DNNs on the augmented datasets is proportional to the amount of the IPK original and augmented images. Overall, this study indicates a superior performance of attention mechanisms-based deep learning models in detecting small and subtle features of grain spikes.

Návaznosti

EF16_026/0008446, projekt VaV

Název: Integrace signálu a epigenetické reprogramování pro produktivitu rostlin

Citovat

ULLAH, Sajid, Klára PANZAROVÁ, Martin TRTÍLEK, Matej LEXA, Vojtěch MÁČALA, Thomas ALTMANN, Kerstin NEUMANN, Jan HEJÁTKO, Markéta PERNISOVÁ a Evgeny GLADILIN. High-throughput spike detection in greenhouse cultivated grain crops with attention mechanisms based deep learning models. Plant Phenomics. AAAS, 2024, roč. 2024, March, s. 1-11. ISSN 2643-6515. Dostupné z: https://dx.doi.org/10.34133/plantphenomics.0155.

@article{2380097,
   author = {Ullah, Sajid and Panzarová, Klára and Trtílek, Martin and Lexa, Matej and Máčala, Vojtěch and Altmann, Thomas and Neumann, Kerstin and Hejátko, Jan and Pernisová, Markéta and Gladilin, Evgeny},
   article_number = {March},
   doi = {http://dx.doi.org/10.34133/plantphenomics.0155},
   keywords = {spike detection; high-throughput image analysis; Attention networks; Deep Neural Networks},
   language = {eng},
   issn = {2643-6515},
   journal = {Plant Phenomics},
   title = {High-throughput spike detection in greenhouse cultivated grain crops with attention mechanisms based deep learning models},
   url = {https://spj.science.org/doi/10.34133/plantphenomics.0155},
   volume = {2024},
   year = {2024}
}

TY  - JOUR
ID  - 2380097
AU  - Ullah, Sajid - Panzarová, Klára - Trtílek, Martin - Lexa, Matej - Máčala, Vojtěch - Altmann, Thomas - Neumann, Kerstin - Hejátko, Jan - Pernisová, Markéta - Gladilin, Evgeny
PY  - 2024
TI  - High-throughput spike detection in greenhouse cultivated grain crops with attention mechanisms based deep learning models
JF  - Plant Phenomics
VL  - 2024
IS  - March
SP  - 1-11
EP  - 1-11
PB  - AAAS
SN  - 26436515
KW  - spike detection
KW  - high-throughput image analysis
KW  - Attention networks
KW  - Deep Neural Networks
UR  - https://spj.science.org/doi/10.34133/plantphenomics.0155
N2  - Detection of spikes is the first important step towards image-based quantitative assessment of crop yield. However, spikes of grain plants occupy only a tiny fraction of the image area and often emerge in the middle of the mass of plant leaves that exhibit similar colors as spike regions. Consequently, accurate detection of grain spikes renders, in general, a non-trivial task even for advanced, state-of-the-art deep learning neural networks (DNNs). To improve pattern detection in spikes, we propose architectural changes to Faster-RCNN (FRCNN) by reducing feature extraction layers and introducing a global attention module. The performance of our extended FRCNN-A vs. conventional FRCNN was compared on images of different European wheat cultivars, including ’difficult’ bushy phenotypes from two different phenotyping facilities and optical setups. Our experimental results show that introduced architectural adaptations in FRCNN-A helped to improve spike detection accuracy in inner regions. The mAP of FRCNN and FRCNN-A on inner spikes is 76.0% and 81.0%, respectively, while on the state-of-the-art detection DNNs, Swin Transformer mAP is 83.0%. As a lightweight network, FRCNN-A is faster than FRCNN and Swin Transformer on both baseline and augmented training datasets. On the FastGAN augmented dataset, FRCNN achieved mAP of 84.24%, FRCNN-A 85.0%, and the Swin Transformer 89.45%. The increase in mAP of DNNs on the augmented datasets is proportional to the amount of the IPK original and augmented images. Overall, this study indicates a superior performance of attention mechanisms-based deep learning models in detecting small and subtle features of grain spikes.
ER  -

ULLAH, Sajid, Klára PANZAROVÁ, Martin TRTÍLEK, Matej LEXA, Vojtěch MÁČALA, Thomas ALTMANN, Kerstin NEUMANN, Jan HEJÁTKO, Markéta PERNISOVÁ a Evgeny GLADILIN. High-throughput spike detection in greenhouse cultivated grain crops with attention mechanisms based deep learning models. \textit{Plant Phenomics}. AAAS, 2024, roč.~2024, March, s.~1-11. ISSN~2643-6515. Dostupné z: https://dx.doi.org/10.34133/plantphenomics.0155.

Podrobný výpis o publikaci