FUSION
FUnctionality Sharing In Open eNvironments
Heinz Nixdorf Chair for Distributed Information Systems
 
Navigation

PhenoDeep: A Deep Learning-Based Approach for Detecting Reproductive Organs from Digitized Herbarium Specimen Images

Title: PhenoDeep: A Deep Learning-Based Approach for Detecting Reproductive Organs from Digitized Herbarium Specimen Images
Authors: Abdelaziz Triki, Bassem Bouaziz, Jitendra Gaikwad, and Walid Mahdi
Source: Proceedings of the International Conference on Neural Information Processing (ICONIP 2021): Lecture Notes in Computer Science
Date: 2021-12-06
Type: Conference Paper
Abstract:

Phenology is an important factor in studying climate change’s effect on plant growth. Recent studies on herbarium specimens have afforded valuable information on plant phenology. The initiatives of herbaria to digitize their collections can extend plant phenological research rapidly by providing online access to significant collections of digitized specimen images. However, they present a major outstanding challenge when extracting reliable data from the specimen sheets. To effectively detect the presence/absence of the reproductive organs such as buds, flowers, and fruits from the specimen images, we developed PhenoDeep, a deep learning approach based on the refined Mask Scoring R-CNN approach. The Mask Scoring R-CNN backbone network was modified by exploiting the advantages of combining ResNet and DenseNet architectures. The experimental results indicate that PhenoDeep can segment the reproductive organs within different specimens, where the precision of PhenoDeep reached 94.1% and recall 94.3%.

URL: https://link.springer.com/chapter/10.1007/978-3-030-92185-9_33
BibTex:
@InProceedings{10.1007/978-3-030-92185-9_33,
author="Triki, Abdelaziz
and Bouaziz, Bassem
and Gaikwad, Jitendra
and Mahdi, Walid",
editor="Mantoro, Teddy
and Lee, Minho
and Ayu, Media Anugerah
and Wong, Kok Wai
and Hidayanto, Achmad Nizar",
title="PhenoDeep: A Deep Learning-Based Approach for Detecting Reproductive Organs from Digitized Herbarium Specimen Images",
booktitle="Neural Information Processing",
year="2021",
publisher="Springer International Publishing",
address="Cham",
pages="402--413",
abstract="Phenology is an important factor in studying climate change's effect on plant growth. Recent studies on herbarium specimens have afforded valuable information on plant phenology. The initiatives of herbaria to digitize their collections can extend plant phenological research rapidly by providing online access to significant collections of digitized specimen images. However, they present a major outstanding challenge when extracting reliable data from the specimen sheets. To effectively detect the presence/absence of the reproductive organs such as buds, flowers, and fruits from the specimen images, we developed PhenoDeep, a deep learning approach based on the refined Mask Scoring R-CNN approach. The Mask Scoring R-CNN backbone network was modified by exploiting the advantages of combining ResNet and DenseNet architectures. The experimental results indicate that PhenoDeep can segment the reproductive organs within different specimens, where the precision of PhenoDeep reached 94.1{\%} and recall 94.3{\%}.",
isbn="978-3-030-92185-9"
}