An Improved 2D Pose Estimation Algorithm for Extracting Phenotypic Parameters of Tomato Plants in Complex Backgrounds DOI Creative Commons
Yawen Cheng,

Ni Ren,

Anqi Hu

et al.

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(23), P. 4385 - 4385

Published: Nov. 24, 2024

Phenotypic traits, such as plant height, internode length, and node count, are essential indicators of the growth status tomato plants, carrying significant implications for research on genetic breeding cultivation management. Deep learning algorithms object detection segmentation have been widely utilized to extract phenotypic parameters. However, segmentation-based methods labor-intensive due their requirement extensive annotation during training, while approaches exhibit limitations in capturing intricate structural features. To achieve real-time, efficient, precise extraction traits seedling tomatoes, a novel phenotyping approach based 2D pose estimation was proposed. We enhanced heatmap-free method, YOLOv8s-pose, by integrating Convolutional Block Attention Module (CBAM) Content-Aware ReAssembly FEatures (CARAFE), develop an improved YOLOv8s-pose (IYOLOv8s-pose) model, which efficiently focuses salient image features with minimal parameter overhead achieving superior recognition performance complex backgrounds. IYOLOv8s-pose manifested considerable enhancement detecting bending points stem nodes. Particularly detection, attained Precision 99.8%, exhibiting improvement over RTMPose-s, YOLOv5s6-pose, YOLOv7s-pose, 2.9%, 5.4%, 3.5%, respectively. Regarding height estimation, achieved RMSE 0.48 cm rRMSE 2%, 65.1%, 68.1%, 65.6%, 51.1% reduction compared When confronted more also exhibited 15.5%, 23.9%, 27.2%, 12.5% YOLOv8s-pose. achieves high precision simultaneously enhancing efficiency convenience, rendering it particularly well suited extracting parameters plants grown naturally within greenhouse environments. This innovative provides new means rapid, intelligent, real-time acquisition

Language: Английский

3D time-series phenotyping of lettuce in greenhouses DOI
Hanyu Ma, Weiliang Wen,

Wenbo Gou

et al.

Biosystems Engineering, Journal Year: 2025, Volume and Issue: 250, P. 250 - 269

Published: Jan. 23, 2025

Language: Английский

Citations

0
Three-Dimensional Reconstruction, Phenotypic Traits Extraction, and Yield Estimation of Shiitake Mushrooms Based on Structure from Motion and Multi-View Stereo DOI Creative Commons

Xingmei Xu,

Jiayuan Li, Jing Zhou

et al.

Agriculture, Journal Year: 2025, Volume and Issue: 15(3), P. 298 - 298

Published: Jan. 30, 2025

Phenotypic traits of fungi and their automated extraction are crucial for evaluating genetic diversity, breeding new varieties, estimating yield. However, research on the high-throughput, rapid, non-destructive fungal phenotypic using 3D point clouds remains limited. In this study, a smart phone is used to capture multi-view images shiitake mushrooms (Lentinula edodes) from three different heights angles, employing YOLOv8x model segment primary image regions. The segmented were reconstructed in Structure Motion (SfM) Multi-View Stereo (MVS). To automatically individual mushroom instances, we developed CP-PointNet++ network integrated with clustering methods, achieving an overall accuracy (OA) 97.45% segmentation. computed phenotype correlated strongly manual measurements, yielding R2 > 0.8 nRMSE < 0.09 pileus transverse longitudinal diameters, = 0.53 RMSE 3.26 mm height, 0.79 0.12 stipe diameter, 0.65 4.98 height. Using these parameters, yield estimation was performed PLSR, SVR, RF, GRNN machine learning models, demonstrating superior performance (R2 0.91). This approach also adaptable extracting other fungi, providing valuable support initiatives.

Language: Английский

Citations

0
Recognition and phenotypic detection of maize stem and leaf at seedling stage based on 3D reconstruction technique DOI

Haiou Guan,

Xueyan Zhang, Xiaodan Ma

et al.

Optics & Laser Technology, Journal Year: 2025, Volume and Issue: 187, P. 112787 - 112787

Published: March 14, 2025

Language: Английский

Citations

0
Cotton3DGaussians: Multiview 3D Gaussian Splatting for boll mapping and plant architecture analysis DOI
Lizhi Jiang, J. F. Sun, Peng W. Chee

et al.

Computers and Electronics in Agriculture, Journal Year: 2025, Volume and Issue: 234, P. 110293 - 110293

Published: March 24, 2025

Language: Английский

Citations

0
An integrated method for phenotypic analysis of wheat based on multi-view image sequences: from seedling to grain filling stages DOI Creative Commons
Shengxuan Sun,

Yeping Zhu,

Shengping Liu

et al.

Frontiers in Plant Science, Journal Year: 2024, Volume and Issue: 15

Published: Aug. 19, 2024

Wheat exhibits complex characteristics during its growth, such as extensive tillering, slender and soft leaves, severe organ cross-obscuration, posing a considerable challenge in full-cycle phenotypic monitoring. To address this, this study presents synthesized method based on SFM-MVS (Structure-from-Motion, Multi-View Stereo) processing for handling segmenting wheat point clouds, covering the entire growth cycle from seedling to grain filling stages. First, multi-view image acquisition platform was constructed capture sequences of plants, dense clouds were generated using technology. High-quality produced by implementing improved Euclidean clustering combined with centroids, color filtering, statistical filtering methods. Subsequently, segmentation plant stems leaves performed region algorithm. Although performance suboptimal jointing, booting stages due glut overlap, there salient improvement leaf efficiency over cycle. Finally, parameters analyzed across different stages, comparing automated measurements height, length, width actual measurements. The results demonstrated coefficients determination ( R2 ) 0.9979, 0.9977, 0.995; root mean square errors (RMSE) 1.0773 cm, 0.2612 0.0335 cm; relative (RRMSE) 2.1858%, 1.7483%, 2.8462%, respectively. These validate reliability accuracy our proposed workflow automatically extracting width, indicating that 3D reconstructed model achieves high precision can quickly, accurately, non-destructively extract parameters. Additionally, convex hull volume, surface area, Crown area extracted, providing detailed analysis dynamic changes throughout ANOVA conducted cultivars, accurately revealing significant differences at various This proposes convenient, rapid, quantitative method, offering crucial technical support dynamics monitoring, applicable precise monitoring wheat.

Language: Английский

Citations

2
PanicleNeRF: low-cost, high-precision in-field phenotyping of rice panicles with smartphone DOI Creative Commons
Xin Yang,

Xuqi Lu,

Pengyao Xie

et al.

Plant Phenomics, Journal Year: 2024, Volume and Issue: 6

Published: Jan. 1, 2024

The rice panicle traits substantially influence grain yield, making them a primary target for phenotyping studies. However, most existing techniques are limited to controlled indoor environments and have difficulty in capturing the under natural growth conditions. Here, we developed PanicleNeRF, novel method that enables high-precision low-cost reconstruction of three-dimensional (3D) models field based on video acquired by smartphone. proposed combined large model Segment Anything Model (SAM) small You Only Look Once version 8 (YOLOv8) achieve segmentation images. neural radiance fields (NeRF) technique was then employed 3D using images with 2D segmentation. Finally, resulting point clouds processed successfully extract traits. results show PanicleNeRF effectively addressed image task, achieving mean F1 score 86.9% Intersection over Union (IoU) 79.8%, nearly double boundary overlap (BO) performance compared YOLOv8. As cloud quality, significantly outperformed traditional SfM-MVS (structure-from-motion multi-view stereo) methods, such as COLMAP Metashape. length accurately extracted rRMSE 2.94% indica 1.75% japonica rice. volume estimated from strongly correlated number ( R 2 = 0.85 0.82 ) mass (0.80 0.76 ). This provides solution high-throughput in-field panicles, accelerating efficiency breeding.

Language: Английский

Citations

2
Prediction of Useful Eggplant Seedling Transplants Using Multi-View Images DOI Creative Commons
Xiangyang Yuan, Jingyan Liu,

Huanyue Wang

et al.

Agronomy, Journal Year: 2024, Volume and Issue: 14(9), P. 2016 - 2016

Published: Sept. 4, 2024

Traditional deep learning methods employing 2D images can only classify healthy and unhealthy seedlings; consequently, this study proposes a method by which to further seedlings into primary secondary finally differentiate three classes of seedling through 3D point cloud for the detection useful eggplant transplants. Initially, RGB types substrate-cultivated (primary, secondary, unhealthy) were collected, classified using ResNet50, VGG16, MobilNetV2. Subsequently, was generated types, series filtering processes (fast Euclidean clustering, filtering, voxel filtering) employed remove noise. Parameters (number leaves, plant height, stem diameter) extracted from found be highly correlated with manually measured values. The box plot shows that clearly differentiated parameters. clouds ultimately directly classification models PointNet++, dynamic graph convolutional neural network (DGCNN), PointConv, in addition complementary operation plants missing leaves. PointConv model demonstrated best performance, an average accuracy, precision, recall 95.83, 95.88%, respectively, loss 0.01. This employs spatial feature information analyse different categories more effectively than two-dimensional (2D) image three-dimensional (3D) extraction methods. However, there is paucity studies applying predict Consequently, has potential identify high accuracy. Furthermore, it enables quality inspection during agricultural production.

Language: Английский

Citations

1
RGB camera-based monocular stereo vision applied in plant phenotype: A survey DOI
Hua Yin, Shan Luo,

Jianjun Tang

et al.

Computers and Electronics in Agriculture, Journal Year: 2024, Volume and Issue: 227, P. 109523 - 109523

Published: Oct. 9, 2024

Language: Английский

Citations

1
Research Progress of Spectral Imaging Techniques in Plant Phenotype Studies DOI Creative Commons
Qian Zhang,

RuPeng Luan,

Ming Wang

et al.

Plants, Journal Year: 2024, Volume and Issue: 13(21), P. 3088 - 3088

Published: Nov. 2, 2024

Spectral imaging technique has been widely applied in plant phenotype analysis to improve trait selection and genetic advantages. The latest developments applications of various optical techniques phenotypes were reviewed, their advantages applicability compared. X-ray computed tomography (X-ray CT) light detection ranging (LiDAR) are more suitable for the three-dimensional reconstruction surfaces, tissues, organs. Chlorophyll fluorescence (ChlF) thermal (TI) can be used measure physiological characteristics plants. Specific symptoms caused by nutrient deficiency detected hyperspectral multispectral imaging, LiDAR, ChlF. Future research based on spectral closely integrated with processes. It effectively support related disciplines, such as metabolomics genomics, focus micro-scale activities, oxygen transport intercellular chlorophyll transmission.

Language: Английский

Citations

1
Soybean seed pest damage detection method based on spatial frequency domain imaging combined with RL-SVM DOI Creative Commons
Xuanyu Chen, Wei He, Zhihao Ye

et al.

Plant Methods, Journal Year: 2024, Volume and Issue: 20(1)

Published: Aug. 20, 2024

Soybean seeds are susceptible to damage from the Riptortus pedestris, which is a significant factor affecting quality of soybean seeds. Currently, manual screening methods for limited visual inspection, making it difficult identify that phenotypically defect-free but have been punctured by stink bugs on sub-surface. To facilitate convenient and efficient identification healthy seeds, this paper proposes seed pest detection method based spatial frequency domain imaging combined with RL-SVM. Firstly, optical data obtained using single integration sphere technique, vigor index through germination experiments. Then, above two items feature extraction algorithms (the successive projections algorithm competitive adaptive reweighted sampling algorithm), characteristic wavelengths soybeans identified. Subsequently, technique used obtain sub-surface images in forward manner, coefficients such as reduced scattering coefficient $${{\mu }{\prime}}_{s}$$ absorption $${\mu }_{a}$$ inverted. Finally, RL-MLR, RL-GRNN, RL-SVM prediction models established ratio area insect-damaged entire seed, varieties, at three (502 nm, 813 712 nm) predicting identifying stinging sucking levels The experimental results show yields small errors less than 15% 10% . After parameter adjustment reinforcement learning, Macro-Recall metrics each model improved 10%-15%, achieves high value 0.9635 classifying

Language: Английский

Citations

0