Machine learning methods for efficient and automated in situ monitoring of peach flowering phenology
Introduction
Plant phenology is the seasonal calendar of plant biological events, including sprouting, shoot development, inflorescence emergence, flowering, fruit development, and senescence (Jochner and Menzel, 2015). Accurate phenology of plants is crucial for agricultural production and operations because it guides specific schedules for fertilizing, irrigation, pest control, harvesting, and breeding (Fu et al., 2019, Piao et al., 2019). Therefore, updated descriptions of crop growth stages provide universal, and standardized phenological scales for major crop species (Hill et al., 2015). The methods for collecting and compiling phenological scales are becoming standardized and objective (d’Andrimont et al., 2020).
The Biologische Bundesanstalt, Bundessortenamt und CHemische Industrie (BBCH) scale, developed by the Federal Biological Research Centre for Agriculture and Forestry (Meier, 2001), outperforms many plant phenological scales and is considered a standard plant phenological scale criterion in plant phenology studies (Stucky et al., 2018). The BBCH method describes plant phenological stages as a decimal system with a two-digit code, from zero to nine. The first digit defines the principal stage, such as leaf development and flowering. In contrast, the second digit denotes the secondary stage, including the degree of development or characteristic changes of a major stage. The BBCH scale has been extended for specific crops, each having specific coding systems. In fruit trees, pome fruit, stone fruit, currants (Meier et al., 1994), citrus (Agustí et al., 1995), and mango (Hernández Delgado et al., 2011), have specific BBCH scales.
Peach is botanically classified as stone fruit. Peach flowering involves two of the eight stone fruit phenological stages in the BBCH scale. The peach flowering stages are inflorescence emergence and flowering, BBCH principal stage codes 5 and 6 (Meier, 2001). Genotype and phenotype differences between early-blossom and late-blossom peaches are significant for the peach industry because blossom timing is linked to fruit maturity timing (Oussama et al., 2008). In south China, fruits of early-blossom peach cultivars tend to mature late. Many peach orchards cultivate two to five cultivars with different fruit maturity timings to elongate the sales season. Therefore, knowing precise peach flowering phenological stages facilitates accurate fruit harvest timing and scheduling harvest (Sánchez-Pérez et al., 2014).
Peach flowers are self-compatible with very short pollination receptive periods. Pollinators such as honey bees facilitate natural pollination for fruit production during the receptive period (Penso et al., 2020). However, when cross-breeding between cultivars with different flowering timings, artificial pollination is conducted during the receptive period (Larue et al., 2021). Male anthers are collected and dried to extract pollen, and female flowers are emasculated before petal opening to prevent natural pollination from other pollen. Additionally, the change of flowering phenology reflects potential heat stress the trees are suffering, which is useful information to make precise irrigation schedules (Vanalli et al., 2021). Appropriate applying of techniques of heat stress management requires comprehensive understanding of phenological changes and the corresponding heat tolerance features (Akter and Rafiqul Islam, 2017). Thus, the precise knowledge of flowering phenological stages is essential for scheduling artificial pollination and precise irrigation. Precise and efficient monitoring methods for peach flowering phenological stages must be developed rather than the current arbitrary manual observations (Prudencio et al., 2018).
Manual observation and determination of peach phenological stages is laborious and empirical; hence, artificial estimates of pollination and harvest timing are inevitable. Several studies have monitored and predicted flowering phenological stages in many plants, including apple, pear, and soybean. For strawberry, a combination of networked microsensors, weather models, and machine learning predicted flowering phenological stages and fruit yield (Lee et al., 2020). Deep learning methods detected apple flower bloom intensity and rice phenology from in-field images using convolutional neural networks to monitor flowering phenological stages (Dias et al., 2018, Han et al., 2021). A similar machine learning method detected and monitored pear flowering phenological stages from pear tree images (Sun et al., 2021). A simple linear iterative clustering (SLIC)-based machine learning method monitored and predicted the soybean flowering timing (So et al., 2017). All the above studies involved machine learning using crop flowering images. These models performed well in different crops and contexts because they were built with various statistical and mathematical principles using training data rather than being explicitly designed for specific tasks (Zhu et al., 2022, Agudelo-Rodríguez et al., 2020). Peach flowers have distinct colors; hence, machine learning based on flower colors is applicable for estimating and monitoring peach flowering phenological stages.
This study analyzed feature differences in peach flower images from several flowering phenological stages. The performances of various machine learning methods (random forest, support vector machine, naïve Bayes, and k-Nearest Neighbors) for estimating and monitoring peach flowering phenological stages were tested using peach flower images and meteorological data. Then, the most optimal and accurate method for automated monitoring of peach flowering phenology is proposed, and the effects of input variables on monitoring accuracy are discussed. Finally, the perspectives and remarks of optimal methods are presented.
Section snippets
Peach trees and study location
Peach cultivars YL (Prunus persica L. Batsch ‘yulu’), CY (Prunus persica L. Batsch ‘chiyue’), and HJYL (Prunus persica L. Batsch ‘hujingyulu’) were selected as early-blossom, mid-blossom, and late-blossom cultivars, respectively. These cultivars are widely cultivated in the south China region. Three peach cultivars were grown in different locations at the resource collection orchard of Fenghua Peach Research Institute (29.687907 N, 121.274495 E). Initially, the peach trees were planted in
Peach flowering phenology and image characteristics
At stages 54 and 55, inflorescences are still closed and covered with green to brown colored scales, but flower buds are visible at stage 55. At stage 57, flower buds are still closed with white or pink visible petals, and stage 59 has a single blooming flower. At stage 63, at least 30 % of the flowers are open. However, at stage 65, over 50 % of the flowers are open, and petals start falling. At stage 67, most petals will have fallen, leaving pistils and stamens, with simultaneous leaf
Discussion
This study presented the potential of machine learning coupled with peach flower images and meteorological data for classifying peach flowering phenological stages among different peach cultivars. The RF method accurately and rapidly identified the BBCH flowering stages of different peach cultivars. The most useful variables in the RF model are DAB, GDD and area proportions of image regions with specific hue values. The RF method applies to peach producers and breeders who require accurate
Conclusions
The automatic monitoring of peach flowering phenology has considerable potential to optimize orchard management and decrease labor costs. This study divided peach flowering phenology into eight stages, from inflorescence emergence (54, 55, 57 and 59) to flowering (63, 65, 67 and 69), following the BBCH scales. A random forest model was trained and tested to classify peach flowering phenological stages using in situ images and meteorological data of peach flowering. The RF model is the most
CRediT authorship contribution statement
Yihang Zhu: Writing – original draft, Software, Validation. Miaojin Chen: Data curation. Qing Gu: Investigation, Supervision, Writing – review & editing. Yiying Zhao: Validation, Data curation. Xiaobin Zhang: Conceptualization, Methodology, Software. Qinan Sun: Validation. Xianbin Gu: Validation. Kefeng Zheng: Supervision.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
The work was supported by the Ningbo Municipal Program of Germplasm Garden for Fenghua Peach; the Key Scientific and Technological Grant of Zhejiang for Breeding New Agricultural Varieties [2021C02066-4-3]; and the 2022 Key Research Program of Zhejiang Academy of Agricultural Sciences: Crop Phenotype High-throughput Acquisition and Intelligent Application Platform [10412030022CC2201G].
References (52)
- et al.
Evaluation of a predictive model to configure an active packaging with moisture adsorption for fresh tomato
Food Packag. Shelf Life
(2020) - et al.
Simple remote sensing detection of Corymbia calophylla flowers using common 3 –band imaging sensors
Remote Sens. Appl.: Soc. Environ.
(2018) - et al.
Detecting flowering phenology in oil seed rape parcels with Sentinel-1 and -2 time series
Rem. Sens. Environ.
(2020) - et al.
Apple flower detection using deep convolutional networks
Comput. Ind.
(2018) - et al.
Satellite prediction of forest flowering phenology
Remote Sens. Environ.
(2021) - et al.
Flower development in sweet cherry framed in the BBCH scale
Sci. Hortic.-Amsterdam
(2015) - et al.
Early detection of tomato spotted wilt virus infection in tobacco using the hyperspectral imaging technique and machine learning algorithms
Comput. Electron. Agr.
(2019) - et al.
Integrated phenology and climate in rice yields prediction using machine learning methods
Ecol. Indic.
(2021) - et al.
Phenological growth stages of mango (Mangifera indica L.) according to the BBCH scale
Sci. Hortic.-Amsterdam
(2011) - et al.
Detection of QTL for metabolic and agronomic traits in wheat with adjustments for variation at genetic loci that affect plant phenology
Plant Sci.
(2015)
Chilling and heat requirement of peach cultivars and changes in chilling accumulation spectrums based on 100-year records in Republic of Korea
Agr. Forest Meteorol.
Efficient monitoring of phenology in chestnuts
Sci. Hortic.-Amsterdam
A framework for predicting soft-fruit yields and phenology using embedded, networked microsensors, coupled weather models and machine-learning techniques
Comput. Electron. Agr.
Comparison of methods for estimation of chilling and heat requirements of nectarine and peach genotypes for flowering
Sci. Hortic.-Amsterdam
Automatic detection and classification of leaf spot disease in sugar beet using deep learning algorithms
Physica A
Codification and description of almond (Prunus dulcis) vegetative and reproductive phenology according to the extended BBCH scale
Sci. Hortic.-Amsterdam
Apple, peach, and pear flower detection using semantic segmentation network and shape constraint level set
Comput. Electron. Agr.
Shifts in the thermal niche of fruit trees under climate change: The case of peach cultivation in France
Agr. Forest Meteorol.
A near real-time deep learning approach for detecting rice phenology based on UAV images
Agr. Forest Meteorol.
Escala BBCH para la descripción de los estadios fenológicos del desarrollo de los agrios (Gén. Citrus)
Levante Agrícola
Heat stress effects and management in wheat. A review
Agron. Sustain. Dev.
Monitoring Plant Phenology Using Digital Repeat Photography
Environ. Manage.
Machine learning modeling of plant phenology based on coupling satellite and gridded meteorological dataset
Int. J. Biometeorol.
Weed Classification for Site-Specific Weed Management Using an Automated Stereo Computer-Vision Machine-Learning System in Rice Fields
Plants
The phyllochron of well-watered and water deficit mature peach trees varies with shoot type and vigour
AoB Plants
Daylength helps temperate deciduous trees to leaf-out at the optimal time
Global Change Biol.
Cited by (3)
PhenoNet: A two-stage lightweight deep learning framework for real-time wheat phenophase classification
2024, ISPRS Journal of Photogrammetry and Remote Sensing