Skip to main content

State-of-the-art Technology and Applications in Crop Phenomics, Volume II

139kviews
Editorial10 May 2023
Editorial: State-of-the-art technology and applications in crop phenomics, volume II

High-throughput acquisition and analysis of phenotypic data is crucial for plant breeding, as phenotypes are the language of plants and a way for them to express their growth status. We can understand plants and discover their secrets of life through the phenotypes. The crop phenotyping community has put a lot of effort into collecting, processing, and analyzing phenotypic data, which is increasingly considered an important tool for rapidly advancing genetic gain in breeding programs (Zhao et al., 2019). Various types of data have emerged from diverse phenotyping platforms ranging from lab-scale to field-scale, and as a result, various phenotypic data processing approaches have emerged at this historic moment. By integrating knowledge from life science, optics, artificial intelligence, computer science, and engineering, plant phenomics has developed into a cutting-edge discipline, and considerable progress has been made in both phenotyping facilities and methodologies.To promote the most advanced research progresses in crop phenomics, this special issue is prepared and released, which covers new advances in crop phenomics, including phenotyping platforms, methods, and applications. Although various phenotyping platforms have been developed, there is still a great demand for low-cost phenotyping equipment. Deep learning, as a highly effective technology, is applied to various types of images such as RGB, hyperspectral, and CT images to produce highly accurate phenotypic parameters. Meanwhile, the potential of machine learning approaches has also been demonstrated in phenotypic analysis, such as post-harvest quality control, breeding, plant research, and plant response to environmental stress.

Image-based high-throughput phenotyping provides a powerful tool for capturing diverse images and quantifying morphological and physiological traits. Compared to expensive commercial high-throughput phenotyping platforms, designing and developing a low-cost phenotyping platform provides an alternative for plant phenomics studies. A low-cost fluorescence phenotyping platform consisting of four UV lamps, 12 white lamps, and a digital camera was built to evaluate dynamic infection process of Tobacco mosaic virus-green fluorescent protein in tobacco leaves.The performance was evaluated by comparing fluorescence images with RGB images acquired by the system, and the results showed that green fluorescence changes could be identified by the cheap and simple equipment and the non-destructive detection of TMV-GFP was realized (Ye, et al.). A miniaturized shoot phenotyping platform, MVS-Pheno V2, has been developed to enable 3D reconstruction of individual plants using multi-view stereo techniques. Based on the MVS-Pheno V1 platform designed for low plant shoots, the MVS-Pheno V2 platform consists of four parts. The hardware is miniaturized to occupy less space, while controlled imaging conditions have been established to avoid the effects of light and wind and ensure high-quality images. Additionally, wireless communication and control have been integrated to avoid cable tangling. The data processing system includes 3D point cloud generation, calibration, and 3D phenotype extraction. To evaluate the performance of the platform, three cultivars of wheat shoots at four different growth stages were used in testing (Wu et al.).Roots are an essential part of all vascular plants that grow underground. To study root phenotyping, a high-throughput in situ root phenotyping platform called the RhizoPot platform was designed and developed with integrated hardware and software. The platform consists of four parts. The RhizoPot cultivated seedlings, while RhizoAuto collected in situ root images. A deep learning framework, DeepLabv3++, was used to segment root images, and WinRhizo software was used to obtain dynamic root phenotypes. Root hair phenotypes were analyzed by RhizoComp. This platform allows for the efficient and high-throughput analysis of the dynamic response characteristics of root phenotypes (Zhao et al.). Besides, the platform was used to investigate the effects of high-temperature weather on cotton seedlings' growth dynamics of the above-ground parts and root phenotype by using images taken by this platform (Fan et al). Dowd et al. designed and developed "3D root Mesocosms" to enable the visualization and analysis of full-sized plant root architecture in 3-dimensions. The system uses a growth container with an internal volume of 45 ft 3 (1.27m 3 ) that is suitable for large crop and bioenergy grass root systems. The setup only requires an appropriate studio space and a digital camera, and the 3D visualization is created using a photogrammetric reconstruction pipeline. The system was evaluated on root systems of switchgrass, maize, and sorghum to demonstrate its capability for comparing and studying different species (Dowd et al.).In addition to the external phenotype parameters of crops, the internal structure of crops is also a key focus of breeders. Imaging technologies such as CT imaging have entered the field of phenotype research. For example, a Micro-CT imaging system was developed to non-destructively acquire CT projection images of passion fruit, followed by 3D model reconstruction (Lu et al.). Comparable accuracy to manual operations was achieved regarding the external traits of passion fruit, while more reliable traits were achieved regarding the internal traits.

Regarding the methods used in plant phenomics, machine learning is one of the most widely used algorithms for data analysis. Among these methods, deep learning is essentially a neural network with three or more layers, allowing it to learn from large amounts of data. The advantages of deep learning, such as its efficient data processing capabilities, automatic feature extraction functions, and high prediction accuracy, make it an important tool for crop phenomics data analysis and prediction.Crop phenomics generates massive amounts of image data of various types, including RGB images, hyperspectral images, CT images, and microscope images.Object detection, which recognizes and detects different objects in an image to classify them, is a key component of numerous agricultural tasks and applications, such as object counting and phenotyping (Wosner et al., 2021). Filed wheat spikes and rice panicles are the most important agronomic traits associated with yield. Thus, quick, and accurate detection and counting of spikes and panicles has always been one of the most important scientific research topics, especially the state-of-the-art performance in object detection and counting brought by deep learning has greatly facilitated progress in the spike and panicle detection and counting. A RetinaNet (SpikeRetinaNet) was trained for spike detection (Wen et al.), and several improvements were made to resolve the issues such as multi-scale feature fusing, efficiency, and occlusion problem.The YOLOX (You Only Look Once) series of deep learning models have been shown to be powerful and versatile object detection models, offering high accuracy, speed, and flexibility, which make them an attractive option for use in plant phenomics. One of the YOLOX models, YOLO5, has been demonstrated to perform well in detecting and counting small panicles in field rice images under varying illumination and across different rice accessions, even with large image sizes (Wang et al.). In addition to its use in rice panicle detection and counting, the YOLOX deep learning model has been applied to the detection and identification of mature soybean stem nodes, and has been shown to be an effective tool for extracting stem-related phenotypes of mature soybeans. In a comparative analysis with other algorithms, YOLOX achieved a maximum average accuracy (mAP) of 94.36% in detecting soybean stem nodes. To facilitate the identification of soybean structural features, a direct search algorithm was designed (Guo et al.).In crop phenomics, image segmentation, partitioning each pixel in a given image to provide an accurate representation of the object shapes, have replaced the traditional manual observation and measurement of phenotypic data (Luo et al., 2023). The developed deep learning models, including VGG, FCN, U-Net, SegNet, DeepLab, etc. are often used for conducting pixel-wise segmentation in crop phenomics. In current research topic, roots were automatically segmented from the in situ root images collected by RhizoPot platform by using DeepLab V3+ network, and it was integrated into the RhizoPot system (Zhao et al.).Beyond the RGB images, a semantic segmentation model, the U-Net convolution neural network, was implemented on the CT images to distinguish different tissues in the samples (Lu et al.). Supervised image segmentation using a deep learning model usually needs to provide high-quality training sets, which are very labor-intensive. To address this issue, a label generation method based on digital image processing was designed, and the segmentation results with higher accuracy were picked as labels to be used for training.Vision based classification is also applied in crop phenomics. Aiming at the online real-time identification and classification of tobacco shred types and their actual production in the field, the efficient and accurate identification of different tobacco shred types were urged. An MS-X-ResNet network was constructed by selecting ResNet50 network as the prominent network, and an accuracy of 96.65% was achieved (Niu et al.).

Hyperspectral imaging is a rapidly developing method of crop phenomics because it captures both spectral and spatial information. Abiotic, biotic, and quality traits in crops in indoor and outdoor growing conditions can be detected, and phenotyping traits can be generated from a cellular to landscape scale (Sarić et al., 2022).Meanwhile, machine learning algorithms have become an essential tool for hyperspectral image analysis due to their outstanding prediction power, and consequently, in-depth study was developed in the physiological and biochemical research of crops by combining hyperspectral imaging and machine learning.In terms of maize leaves, estimating the amino acid content can be useful for improving yield. Researchers used the PLSR (Partial Least Square Regression) algorithm to create various models for amino acid content by analyzing the reflectance of all bands, sensitive band ranges, and sensitive bands. The study demonstrated the potential for machine learning and hyperspectral imaging to be combined for genetic sensitivity analysis and variety improvement of maize (Shu et al.) In addition to leaves, seeds are also a focus of phenotypic research using hyperspectral imaging. In the study of wheat seeds, purity was identified as a key factor in the population of hybrid wheat. The transmittance and reflectance spectra provided a better solution for classifying hybrids and female parents than reflectance spectra alone. Specifically, a classification model, Detrend-CARS-PLS-DA based on the PLSR algorithm, was established using the transmittance spectrum combined with a characteristic wavelength-screening algorithm (Zhang et al.).Deep learning models have demonstrated their power in crop phenotyping. Yu et al.provide an exciting strategy for predicting phenotyping traits of lettuces from spectral reflectance by developing two end-to-end models based on 2D CNN and FCNN, highlighting the potential contribution of combining deep learning models with spectroscopy for phenotype trait quantification (Yu et al.). The use of hyperspectral imaging for stress symptom detection has been demonstrated (Lowe et al., 2017). The potential of combining hyperspectral reflectance and deep learning in assessing cotton drought resistance among different genotypes has also been shown. A 1D-CNN was designed and established to screen the spectral information of cotton leaves because it had natural structural consistency with one-dimensional spectral information. A vector with 1024 spectral features was used as the input, and the output layer was the chlorophyll fluorescence parameter Fv/Fm prediction value, which is significant in abiotic plant stress (Guo et al.).

The combination of machine learning and hyperspectral imaging has led to significant achievements, and applying machine learning algorithms to other types of imaging data holds considerable promise. LF-NMR/MRI, a non-destructive and accurate technology for assessing water status, has been widely applied in the fields of food The papers in this research topic highlight new advances in phenotyping facilities, methodologies, and analysis that enable the collection of diverse image data ranging from the micro level to the whole-plant level. Phenotyping methodologies and phenotypic analyses, which range from the cellular level to the field scale, utilize various machine learning/deep learning and computer vision algorithms. These papers provide valuable insights into how advanced technologies can be used to measure and analyze crop traits and how this information can be used to improve crop breeding, management, and response to environmental stress. They demonstrate the potential of cutting-edge technologies to revolutionize crop research and production.

WY initially drafted the manuscript with inputs from other topic editors. All the topic editors read and approved the final manuscript.

2,133 views

4 citations

METHODS
12 January 2023
Nondestructive 3D phenotyping method of passion fruit based on X-ray micro-computed tomography and deep learning
Yuwei Lu, 10 more and Qian Liu
Nondestructive 3D phenotyping method of passion fruit based on X-ray micro-computed tomography and deep learning
Passion fruit is a tropical liana of the Passiflora family that is commonly planted throughout the world due to its abundance of nutrients and industrial value. Researchers are committed to exploring the relationship between phenotype and genotype to promote the improvement of passion fruit varieties. However, the traditional manual phenotyping methods have shortcomings in accuracy, objectivity, and measurement efficiency when obtaining large quantities of personal data on passion fruit, especially internal organization data. This study selected samples of passion fruit from three widely grown cultivars, which differed significantly in fruit shape, size, and other morphological traits. A Micro-CT system was developed to perform fully automated nondestructive imaging of the samples to obtain 3D models of passion fruit. A designed label generation method and segmentation method based on U-Net model were used to distinguish different tissues in the samples. Finally, fourteen traits, including fruit volume, surface area, length and width, sarcocarp volume, per-icarp thickness, and traits of fruit type, were automatically calculated. The experimental results show that the segmentation accuracy of the deep learning model reaches more than 0.95, the measurement accuracy of external traits of passion fruit is comparable to manual operations, and the measurement of internal traits is more reliable because of the nondestructive characteristics of our method. According to the statistical data of the whole samples, the Pearson analysis method was used to ana-lyze the possible correlation between the various traits. The principal component analysis was used to evaluate the comprehensive quality of the passion fruit samples. The results of this study will firstly provide a nondestructive method for more accu-rate and efficient automatic acquisition of comprehensive phenotypic traits of pas-sion fruit and have the potential to be extended to more fruit crops. The preliminary study of the correlation between the characteristics of passion fruit can also provide a particular reference value for molecular breeding and comprehensive quality evaluation.

4,639 views

36 citations

ORIGINAL RESEARCH
15 November 2022
Machine learning assisted dynamic phenotypes and genomic variants help understand the ecotype divergence in rapeseed
Hui Feng, 9 more and Wanneng Yang

2,376 views

2 citations

ORIGINAL RESEARCH
8 November 2022
Response of root and root hair phenotypes of cotton seedlings under high temperature revealed with RhizoPot
Cong Fan, 8 more and Yongjiang Zhang

4,803 views

26 citations

ORIGINAL RESEARCH
18 October 2022
Predicting Fv/Fm and evaluating cotton drought tolerance using hyperspectral and 1D-CNN
Congcong Guo, 10 more and Cundong Li

8,942 views

42 citations

RhizoPot platform: A high-throughput in situ root phenotyping platform with integrated hardware and software
METHODS
29 September 2022
RhizoPot platform: A high-throughput in situ root phenotyping platform with integrated hardware and software
Hongjuan Zhao, 11 more and Cundong Li
Quantitative analysis of root development is becoming a preferred option in assessing the function of hidden underground roots, especially in studying resistance to abiotic stresses. It can be enhanced by acquiring non-destructive phenotypic information on roots, such as rhizotrons. However, it is challenging to develop high-throughput phenotyping equipment for acquiring and analyzing in situ root images of root development. In this study, RhizoPot platform, a high-throughput in situ root phenotyping platform integrating plant culture, automatic in situ root image acquisition, and image segmentation, is proposed for quantitative analysis of root development. About one to five plants (maximum plant height; 1.0 meters) can simultaneously grow in each RhizoPot, and the longest growth period depends on the type of plant and environmental conditions. The imaging control software (RhizoAuto) can automatically and non-destructively image the roots of RhizoPot-cultured plants based on the set time and resolution (50-4800 dpi) and obtain high-resolution (>1200 dpi) images in batches. The improved DeepLabv3+ tool was used for batch processing of root images. The roots were automatically segmented and extracted from the background for analysis of information on radical features using conventional root software (WinRhizo and RhizoVision Explorer). Root morphology, root growth rate, and lifespan analysis were conducted using in situ root images and segmented images. The platform illustrates the dynamic response characteristics of root phenotypes in cotton. In conclusion, the RhizoPot platform has the characteristics of low cost, high efficiency, and high throughput. Therefore, it can effectively monitor the development of plant roots and realize the quantitative analysis of root phenotypes in situ.

8,776 views

42 citations

Field rice panicle detection and counting based on deep learning
ORIGINAL RESEARCH
12 August 2022
Field rice panicle detection and counting based on deep learning
Xinyi Wang, 6 more and Lingfeng Duan
Panicle number is directly related to rice yield, so panicle detection and counting has always been one of the most important scientific research topics. Panicle counting is a challenging task due to many factors such as high density, high occlusion, and large variation in size, shape, posture et.al. Deep learning provides state-of-the-art performance in object detection and counting. Generally, the large images need to be resized to fit for the video memory. However, small panicles would be missed if the image size of the original field rice image is extremely large. In this paper, we proposed a rice panicle detection and counting method based on deep learning which was especially designed for detecting rice panicles in rice field images with large image size. Different object detectors were compared and YOLOv5 was selected with MAPE of 3.44% and accuracy of 92.77%. Specifically, we proposed a new method for removing repeated detections and proved that the method outperformed the existing NMS methods. The proposed method was proved to be robust and accurate for counting panicles in field rice images of different illumination, rice accessions, and image input size. Also, the proposed method performed well on UAV images. In addition, an open-access and user-friendly web portal was developed for rice researchers to use the proposed method conveniently.

12,498 views

78 citations

ORIGINAL RESEARCH
8 August 2022
A miniaturized phenotyping platform for individual plants using multi-view stereo 3D reconstruction
Sheng Wu, 7 more and Xinyu Guo

9,919 views

92 citations

ORIGINAL RESEARCH
3 August 2022
Estimation of amino acid contents in maize leaves based on hyperspectral imaging
Meiyan Shu, 4 more and Yuntao Ma

4,878 views

18 citations

ORIGINAL RESEARCH
18 August 2022
Tobacco shred varieties classification using Multi-Scale-X-ResNet network and machine vision
Qunfeng Niu, 4 more and Qiang Yuan

5,026 views

16 citations

METHODS
2 September 2022
An automatic fluorescence phenotyping platform to evaluate dynamic infection process of Tobacco mosaic virus-green fluorescent protein in tobacco leaves

4,323 views

12 citations

ORIGINAL RESEARCH
30 June 2022
Hyperspectral Technique Combined With Deep Learning Algorithm for Prediction of Phenotyping Traits in Lettuce
Shuan Yu, 5 more and Chunjiang Zhao
Hyperspectral Technique Combined With Deep Learning Algorithm for Prediction of Phenotyping Traits in Lettuce
The currently available methods for evaluating most biochemical traits of plant phenotyping are destructive and have extremely low throughput. However, hyperspectral techniques can non-destructively obtain the spectral reflectance characteristics of plants, which can provide abundant biophysical and biochemical information. Therefore, plant spectra combined with machine learning algorithms can be used to predict plant phenotyping traits. However, the raw spectral reflectance characteristics contain noise and redundant information, thus can easily affect the robustness of the models developed via multivariate analysis methods. In this study, two end-to-end deep learning models were developed based on 2-D convolutional neural networks (2DCNN) and fully connected neural networks (FCNN) (Deep2D and DeepFC, respectively) to rapidly and non-destructively predict the phenotyping traits of lettuces from spectral reflectance. Three linear and two non-linear multivariate analysis methods were used to develop models to weigh the performance of the deep learning models. The models based on multivariate analysis methods require a series of manual feature extractions, such as pretreatment and wavelength selection, while the proposed models can automatically extract the features in relation to phenotyping traits. A visible near-infrared hyperspectral camera was used to image lettuce plants growing in the field, and the spectra extracted from the images were used to train the network. The proposed models achieved good performance with a determination coefficient of prediction (Rp2) of 0.9030 and 0.8490 using Deep2D for soluble solids content (SSC) and DeepFC for pH, respectively. The performance of the deep learning models was compared with five multivariate analysis method. The quantitative analysis showed that the deep learning models had higher Rp2 than all the multivariate analysis methods, indicating better performance. Also, wavelength selection and different pretreatment methods had different effects on different multivariate analysis methods, and the selection of appropriate multivariate analysis methods and pretreatment methods increased more time and computational cost. Unlike multivariate analysis methods, the proposed deep learning models did not require any pretreatment or dimensionality reduction and thus are more suitable for application in high-throughput plant phenotyping platforms. These results indicate that the deep learning models can better predict phenotyping traits of plants using spectral reflectance.

4,886 views

36 citations

METHODS
18 April 2022
LF-NMR/MRI Determination of Different 6-Benzylaminopurine Concentrations and Their Effects on Soybean Moisture
Ying Gu, 5 more and Bin Luo

3,173 views

14 citations

ORIGINAL RESEARCH
29 March 2022
Non-destructive Storage Time Prediction of Newhall Navel Oranges Based on the Characteristics of Rind Oil Glands
Shumin Gao, 5 more and Shanjun Li

3,017 views

16 citations