Crop Lodging Prediction From UAV-Acquired Images of Wheat and Canola Using a DCNN Augmented With Handcrafted Texture Features

Sara Mardanisamani, Farhad Maleki, Sara Hosseinzadeh Kassani, Sajith Rajapaksa, Hema Sudhakar Duddu, Menglu Wang, Steve Shirtliffe, Seungbum Ryu, Anique Josuttes, Ti Zhang, Sally Vail, Curtis Pozniak, Isobel Parkin, Ian Stavness, Mark G. Eramian

Introduction

In modern precision agriculture, monitoring crop health at scale is essential for optimizing yields and breeding resilient varieties. One of the most significant challenges in cereal and oilseed production is "lodging"—the permanent displacement of crop stems from their vertical position due to environmental factors like wind and rain. Researchers have recently introduced LodgedNet, a lightweight deep learning architecture designed to automate lodging detection from Unmanned Aerial Vehicle (UAV) imagery with high precision and minimal computational overhead.

What Problem Does This Paper Address?

Lodging is more than just a visual irregularity; it can reduce crop yields by up to 50%, compromise seed quality, and significantly complicate mechanical harvesting. For plant breeders managing thousands of experimental plots, manual assessment is labor-intensive, subjective, and prone to error.

Aerial view of experimental breeding plots used for lodging assessment.

While high-resolution drone imagery provides the necessary detail to observe individual plots, processing this data traditionally required either overly simplistic handcrafted features or massive Deep Convolutional Neural Networks (DCNNs) that demand significant hardware resources. The authors sought to bridge this gap by creating a model that is both highly accurate and efficient enough for deployment on field-ready hardware.

Methods: The LodgedNet Architecture

The core innovation of this research is the LodgedNet model, a hybrid architecture that combines the strengths of deep learning with classical computer vision:

  1. DCNN Backbone: A streamlined 7-layer convolutional network that automatically extracts abstract features from raw image data.
  2. Handcrafted Texture Module: A parallel processing track that calculates specific mathematical descriptors of image texture, specifically Local Binary Patterns (LBP) and Gray-Level Co-occurrence Matrix (GLCM) features.
  3. Feature Fusion: The outputs from both the DCNN and the texture module are concatenated, allowing the final classification layers to benefit from both learned and mathematically defined spatial relationships.

Datasets and Experimental Setup

The model was trained and validated using high-resolution orthomosaic images from breeding trials in Saskatoon, Canada. The dataset included:

  • Crops: 465 wheat plots and 1,638 canola plots.
  • Imaging: 5-channel multispectral data (RGB, Near-Infrared, and Red-Edge) at a resolution of 15–26 mm/pixel.
  • Ground Truth: Expert agronomists labeled each plot as "lodged" or "not lodged."

Key Insights: Texture as a Discriminator

The fundamental premise of LodgedNet is that lodging drastically alters the visual texture of a crop canopy. Upright crops present a relatively uniform, vertical "grain," whereas lodged crops create chaotic, flattened patterns.

Visual comparison of upright and lodged wheat (a, b) and canola (c, d) plots.

By explicitly incorporating GLCM and LBP features, the researchers provided the model with a "shortcut" to understanding these textural shifts. This domain-specific knowledge allows the DCNN component to remain shallow and efficient without sacrificing performance.

Results and Efficiency

LodgedNet demonstrated exceptional performance across both crop types:

  • Accuracy: The model achieved 97.70% accuracy for wheat and 99.06% for canola.
  • Parameter Efficiency: With only 332,306 parameters, LodgedNet is approximately 432 times smaller than industry-standard models like VGG19 (143 million parameters) and 3.7 times smaller than SqueezeNet.
  • Inference Speed: The model processes samples in roughly 3 milliseconds on a GPU, supporting real-time agricultural monitoring applications.

Conclusion and Future Directions

LodgedNet proves that "bigger is not always better" in deep learning for specialized industrial tasks. By fusing lightweight DCNNs with handcrafted texture features, the researchers achieved state-of-the-art results with a fraction of the computational cost.

Future work will focus on parallelizing the texture feature extraction on CPUs to eliminate current bottlenecks and exploring additional features like Gabor filters to better detect the specific direction of crop lodging. This approach paves the way for more accessible, high-throughput phenotyping tools for agronomists and breeders worldwide.

Summary updated

Abstract

Lodging, the permanent bending over of food crops, leads to poor plant growth\nand development. Consequently, lodging results in reduced crop quality, lowers\ncrop yield, and makes harvesting difficult. Plant breeders routinely evaluate\nseveral thousand breeding lines, and therefore, automatic lodging detection and\nprediction is of great value aid in selection. In this paper, we propose a deep\nconvolutional neural network (DCNN) architecture for lodging classification\nusing five spectral channel orthomosaic images from canola and wheat breeding\ntrials. Also, using transfer learning, we trained 10 lodging detection models\nusing well-established deep convolutional neural network architectures. Our\nproposed model outperforms the state-of-the-art lodging detection methods in\nthe literature that use only handcrafted features. In comparison to 10 DCNN\nlodging detection models, our proposed model achieves comparable results while\nhaving a substantially lower number of parameters. This makes the proposed\nmodel suitable for applications such as real-time classification using\ninexpensive hardware for high-throughput phenotyping pipelines. The GitHub\nrepository at https://github.com/FarhadMaleki/LodgedNet contains code and\nmodels.\n
BibTeX
@inproceedings{mardanisamani2019crop,
  author = {Sara Mardanisamani and Farhad Maleki and Sara Hosseinzadeh Kassani and Sajith Rajapaksa and Hema Sudhakar Duddu and Menglu Wang and Steve Shirtliffe and Seungbum Ryu and Anique Josuttes and Ti Zhang and Sally Vail and Curtis Pozniak and Isobel Parkin and Ian Stavness and Mark G. Eramian},
  title = {{Crop Lodging Prediction From UAV-Acquired Images of Wheat and Canola Using a DCNN Augmented With Handcrafted Texture Features}},
  booktitle = {Workshops @ CVPR},
  year = {2019},
  doi = {10.1109/cvprw.2019.00322},
}
Figure