??AI for Wheat: Lodging Resistance & Assessment ??

Lodging happens when crop stems bend or fall over due to wind, rain, or their own weight. This reduces crop yield and quality, makes harvesting harder, and can lead to more diseases. Using drones (UAVs) to assess crop lodging can help farmers quickly spot and monitor these issues. Drones take detailed aerial photos that show where and how badly crops are affected.

UAVs-based crop lodging susceptibility assessment is considered one of the most innovative methods for enhancing agricultural management.

Today we will look at how drone-based crop lodging assessment innovatively enhances early detection, precise monitoring, and informed decision-making in wheat farming.

Automated Grading of Wheat Lodging Using Deep Learning

Country: China ????

Published: 23 June 2024

This study aims to develop a deep learning model for automatically grading wheat lodging severity from UAV images using an advanced multitask learning approach.

The research employed a multitask neural network model, MLP_U-Net, integrating multilayer perceptron (MLP) and U-Net architectures. Data were collected using a DJI 4 Pro UAV, producing high-resolution images. These images were preprocessed and split into training and testing sets (487 and 689 plots, respectively). The model's performance was evaluated using metrics like accuracy (ACC), precision (Pre), recall (Rec), F1 score, and intersection over union (IoU). An ablation study was conducted to assess the impact of different modules on model performance.

The results indicated that the MLP_U-Net model achieved high accuracy in grading wheat lodging severity, with grading accuracies of 96.1% and 84.1% for lodging degree and areas, respectively, in two datasets.

The F1 scores for these tasks were 81.3% and 82.0%, demonstrating the model's robustness. The study also found that MLP_U-Net outperformed conventional machine learning methods, reducing manual workload and improving grading consistency.

Agricultural researchers and UAV technology developers can apply these findings to enhance crop monitoring and management.

Main tools/technologies

1. MLP_U-Net neural network
2. DJI 4 Pro UAV
3. PyTorch framework
4. NVIDIA GeForce RTX 3090 GPU
5. Intel Core i7-10600 CPU

Lightweight Network Revolutionizes Wheat Lodging Detection

Country: China ????

Published: 1 February 2024

This study presents an ultra-lightweight deep learning model, Lodging-U2NetP, designed to efficiently segment wheat lodging areas from UAV imagery, addressing limitations of existing large-volume networks.

The research employed the L-U2NetP model, which integrates Dual Cross-Attention (DCA) and Crisscross Attention (CCA) modules for enhanced feature extraction. UAV images were captured using a DJI MAVIC AIR at 30m altitude with a resolution of 243.2 mm2/pixel. Two datasets were created using different annotation strategies: Crop-annotation (CA set) and Annotation-crop (AC set).

Novel Annotation Strategy: The introduction of the Crop-annotation strategy, where images are cropped before annotation. This method enhances annotation efficiency and model performance by focusing on local features and alleviating the computational burden.

The CA set included 10,460 images, while the AC set comprised 3,000 images. The models were trained and tested on these datasets, with performance evaluated using accuracy, F1 score, and Intersection over Union (IoU) metrics.

Results showed that L-U2NetP achieved high segmentation performance with accuracy, F1 score, and IoU values of 95.45%, 93.11%, and 89.15% on the CA simple set, and 89.72%, 79.95%, and 70.24% on the difficult set, respectively. It demonstrated superior robustness in real-time detection simulations under varying conditions such as lighting changes, occlusions, and motion blur.

The "difficult set" in this research comprises test images that present greater challenges for accurate segmentation. These images feature small or scattered lodging areas, varying degrees of lodging, compressed strip-like lodging shapes, and unclear boundaries due to similar colors and textures. Evaluating the model on this difficult set helps assess its robustness and effectiveness in real-world, challenging conditions.

The model also outperformed traditional networks like U2NetP, UNet, and SegNet, particularly in challenging scenarios.

Agricultural researchers and UAV technology developers can leverage these findings to enhance real-time crop monitoring and management.

Main tools/technologies:

1. Lodging-U2NetP neural network
2. DJI MAVIC AIR UAV
3. PyTorch framework
4. NVIDIA GeForce RTX4070 GPU
5. Intel Xeon Gold 6130T CPU

AI Helps to Discover Wheat Lodging Resistance

Country: Iran ????

Published: 17 October 2023

This study investigates the efficiency of machine learning models, specifically Random Forest (RF), in predicting lodging in bread wheat genotypes using various morphological traits.

The research utilized an alpha-lattice experiment design involving 228 wheat accessions grown over two cropping seasons (2018-2019 and 2019-2020). Traits such as plant height, internode lengths, and diameters were measured using image processing techniques on 20 isolated plants per plot. Multiple linear regression (MLR), support vector regression (SVR), artificial neural networks (ANN), and RF models were employed to predict the lodging score index (LS). The dataset was split into training (75%) and testing (25%) subsets to evaluate model performance using metrics like R2 and RMSE.

Key findings revealed that the RF model outperformed others, achieving R2 values of 0.887 and 0.768 for training and testing data, respectively, with corresponding RMSE values of 0.091 and 0.124.

The lodging score index had high positive correlations with plant height (r=0.78) and internode lengths (r=0.70), indicating these traits as significant predictors.

RF's robustness and accuracy in predicting lodging make it a valuable tool for non-destructive monitoring and precise crop management.

Main tools/technologies

• Random Forest (RF) regression
• Canon SX540HS camera
• Python 3.7
• R 4.3.1 software

(A). Division of the plot into four quadrants Q1, Q2, Q3, and Q4 (B). LA1, LA2, LA3, and LA4 are corresponding to the lodged area in each quadrant. In this scenario, H1 and H2 present the healthy subplots while L1 to L6 are the lodged subplots. The CAI is estimated via averaging the CAI and LA calculated in the six lodged subplots and in each quadrant, respectively

Abbreviations: Lodged area or LA (A), crop angle of inclination or CAI (B), lodging score index or LS (C), plant height or PH (D), number of nodes or NFN (E), peduncle length or PL (F), penultimate length or Pel (G), internode length 2 or IL2 (H), internode length 1 or IL1 (I), peduncle diameter or PD (J), penultimate diameter or PeD (K), internode diameter 2 or ID2 (L), internode diameter 1 or ID1 (M), days to heading or DTH (N), days to flowering or DTF (O), days to maturity or DTM (P), spike weight or SW (Q), spike area or SA (R), and grain yield or GY (S)

Lodged area (LA), crop angle of inclination (CAI), lodging score index (LS), plant height (PH), number of nodes (NFN), peduncle length (PL), penultimate length (Pel), internode length 1 (IL1), internode length 2 (IL2), peduncle diameter (PD), penultimate diameter (PeD), internode diameter 1 (ID1), internode diameter 2 (ID2), days to heading (DTH), days to flowering (DTF), days to maturity (DTM), spike weight (SW), spike area (SA) and grain yield (GY)

Variable biplot for the traits (A) and individual biplot for 228 wheat genotypes (B). Lodged area (LA), crop angle of inclination (CAI), lodging score index (LS), plant height (PH), number of nodes (NFN), peduncle length (PL), penultimate length (Pel), internode length 1 (IL1), internode length 2 (IL2), peduncle diameter (PD), penultimate diameter (PeD), internode diameter 1 (ID1), internode diameter 2 (ID2), days to heading (DTH), days to flowering (DTF), days to maturity (DTM), spike weight (SW), spike area (SA) and grain yield (GY)

Predicted and measured lodging score index of wheat accessions using various regression methods: Scatter plot of predicted and measured lodging score index in training and testing stage of MLR (A and A?), ANN (B and B?), SVR (C and C?) and RF (D and D?)

Actual vs prediction values lodging score index by using MLR, ANN, SVR and RF with training (A) and testing (B) datasets with?±?25% error line

Abbreviations: Lodged area (LA), crop angle of inclination (CAI), lodging score index (LS), plant height (PH), number of nodes (NFN), peduncle length (PL), penultimate length (Pel), internode length 1 (IL1), internode length 2 (IL2), peduncle diameter (PD), penultimate diameter (PeD), internode diameter 1 (ID1), internode diameter 2 (ID2), days to heading (DTH), days to flowering (DTF), days to maturity (DTM), spike weight (SW), spike area (SA) and grain yield (GY)

?? What's Next in Wheat Tech?

In the next edition of our newsletter, we can take a look at pest and disease detection in wheat. Alternatively, we can dive deeper into the postharvest phase and assess the quality assurance of wheat using innovative technologies.

?How interesting is any of these topics for you?

Let us know in the comments below ??

??Share with relevant people to spread knowledge around.

Thank you for your time ??

Wishes of lodge-resistant wheat crops,

Maryna Kuzmenko , Chief Wheat Farming Officer at Petiole Pro

Photo credit for the cover image of today's edition

Rabieyan, E., Darvishzadeh, R. & Alipour, H. Identification and estimation of lodging in bread wheat genotypes using machine learning predictive algorithms. Plant Methods 19, 109 (2023). https://doi.org/10.1186/s13007-023-01088-w

References for today's "AI for Wheat: Lodging Resistance & Assessment"