Chenghao Lu

Chenghao Lu

Auto-generated profile, based on publicly available information

Chenghao Lu

Precision Agriculture Lab, School of Life Sciences, Technical University of Munich

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
21 April 2025 in Smart Agricultural Technology
Verified
Journal Article
Open access
CLChenghao Lu
Chenghao Lu
KGKlaus Gehring
Klaus Gehring
SKStefan Kopfinger
Stefan Kopfinger
Heinz BernhardtHeinz Bernhardt
Heinz Bernhardt
MBMichael Beck
Michael Beck
Kang YuKang Yu
Kang Yu

16 August 2024 in bioRxiv
Verified
Preprint
Open access
PDF file
CLChenghao Lu
Chenghao Lu
Kang YuKang Yu
Kang Yu
Weeds significantly impact agricultural production, and traditional weed control methods often harm soil health and the environment. This study aims to develop deep learning-based segmentation models in identifying weeds in potato fields captured by Unmanned Aerial Vehicle (UAV ) orthophotos and to explore the effects of weeds on potato yield. UAVs were used to collect RGB data from potato fields, flying at an altitude of 10m, with Real-ESRGAN Super-Resolution (SR) enhancing image resolution. We applied the Segment Anything Model (SAM) to do semi-automatic annotation, followed by training the YOLOv8 and MASK-RCNN models for segmentation. Also we used ANOVA and linear regression to analyze the effects of weeds and nitrogen fertilizer on yield. Results showed that the detection accuracy mAP50 scores for YOLOv8 and Mask R-CNN were 0.902 and 0.920, respectively, with the Real-ESRGAN-enhanced model achieving 0.909. When multiple weed types were present, accuracy decreased to 0.86. The linear regression model, incorporating plant and weed coverage areas, explained 41.2% of yield variation (R 2 = 0.412, p-value = 0.01). Both YOLOv8 and Mask R-CNN achieved high accuracy, with YOLOv8 converging faster. Real-ESRGAN reconstruction slightly improved accuracy. Different nitrogen fertilizer treatments had no significant effect on yield, while weed control treatments significantly impacted yield, showing the importance of precise weed mapping.. This study provides insights into weed segmentation and contributes to environmentally friendly precision weed control. Graphic

18 September 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
CLChenghao Lu
Chenghao Lu
CLChang Liu
Chang Liu
SPSilvian Poque
Silvian Poque
AAAndrius Aleliunas
Andrius Aleliunas
ENEgle Norkeviciene
Egle Norkeviciene
XSXiang Su
Xiang Su
Timothy (Phleum pratense) is an important Nordic forage crop whose drought-tolerance breeding depends on accurate biomass-yield assessment. High-throughput plant phenotyping (HTPP) generates multimodal time-series data, but Transformer-based temporal integration can be parameter intensive, and unequal observation intervals complicate modelling. Phenotype–yield relationships can also shift across developmental stages, limiting model reuse. We develop LiquidFormer, combining gated multimodal fusion, a compact Closed-form Continuous-time (CfC) backbone, and attention pooling. CfC accounts for elapsed time through closed-form, input-dependent dynamics, whereas Low-Rank Adaptation (LoRA) enables parameter-efficient cross-experiment calibration. In a non-vernalized greenhouse experiment, LiquidFormer achieved competitive predictive performance (MAE = 1.83 g; (R^2 = 0.797) with 47% fewer parameters than the Transformer. For adaptation to a vernalized experiment with a shifted phenotype–yield relationship, rank-2 LoRA updated only 13.0% of the model parameters and achieved performance close to full fine-tuning and training from scratch. An independent unmanned aerial vehicle field trial further supported the feasibility of LiquidFormer for canopy-scale yield prediction. Overall, LiquidFormer provides a lower-parameter approach to continuous-time multimodal phenotyping, while LoRA supports cross-experiment recalibration by updating a restricted subset of model parameters.

12 October 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
FWFei Wu
Fei Wu
SYShuai Yang
Shuai Yang
CLChenghao Lu
Chenghao Lu
FLFei Li
Fei Li
YHYuncai Hu
Yuncai Hu
Kang YuKang Yu
Kang Yu
The size and vertical distribution of maize leaves can affect plant light interception and canopy photosynthetic efficiency. However, Accurate non-destructive estimation of the vertical distribution of maize leaf area (LA) remains a great challenge. To overcome these limitations, this study presents a novel method to estimate the vertical distribution of maize LA under different nitrogen (N) treatments using an optimized bell-shaped function integrated with UAV-derived maximum leaf area (LAmax) data. Two field experiments were conducted to establish the model and for independent validation, respectively. The results demonstrated that UAV-derived plant height can be used to predict LAmax with an R2 of 0.811, and the proposed method can estimate LA across different growth stages (V9, V12, and VT), with training R2 values ranging from 0.799 to 0.923 and test R2 values from 0.780 to 0.837. The model trained on all training data exhibited good performance as well, achieving a training R² of 0.855. Validation using LAI data from a separate experimental field yielded an R² of 0.714, indicating that model predictive capability was beyond the plots used for model development. Thus, our study suggests the potential of integrating UAV-derived structural information with an optimized bell-shaped function to efficiently and non-destructively estimate vertical LA distribution in maize under different N treatments.