[Paper Review] Detecting Volunteer Cotton Plants in a Corn Field with Deep Learning on UAV Remote-Sensing Imagery

Pappu Kumar Yadav, J. Alex Thomasson+10 more|arXiv (Cornell University)|Jul 14, 2022

Forest Insect Ecology and Management4 citations

TL;DR

This study proposes a YOLOv3-based deep learning approach to detect volunteer cotton plants in corn fields using RGB aerial imagery from unmanned aerial vehicles (UAVs). The method achieves consistent mAP across three image scales (320×320, 416×416, 512×512), demonstrating robust real-time potential for automated detection and targeted spraying to prevent pest reinfestation.

ABSTRACT

The cotton boll weevil, Anthonomus grandis Boheman is a serious pest to the U.S. cotton industry that has cost more than 16 billion USD in damages since it entered the United States from Mexico in the late 1800s. This pest has been nearly eradicated; however, southern part of Texas still faces this issue and is always prone to the pest reinfestation each year due to its sub-tropical climate where cotton plants can grow year-round. Volunteer cotton (VC) plants growing in the fields of inter-seasonal crops, like corn, can serve as hosts to these pests once they reach pin-head square stage (5-6 leaf stage) and therefore need to be detected, located, and destroyed or sprayed . In this paper, we present a study to detect VC plants in a corn field using YOLOv3 on three band aerial images collected by unmanned aircraft system (UAS). The two-fold objectives of this paper were : (i) to determine whether YOLOv3 can be used for VC detection in a corn field using RGB (red, green, and blue) aerial images collected by UAS and (ii) to investigate the behavior of YOLOv3 on images at three different scales (320 x 320, S1; 416 x 416, S2; and 512 x 512, S3 pixels) based on average precision (AP), mean average precision (mAP) and F1-score at 95% confidence level. No significant differences existed for mAP among the three scales, while a significant difference was found for AP between S1 and S3 (p = 0.04) and S2 and S3 (p = 0.02). A significant difference was also found for F1-score between S2 and S3 (p = 0.02). The lack of significant differences of mAP at all the three scales indicated that the trained YOLOv3 model can be used on a computer vision-based remotely piloted aerial application system (RPAAS) for VC detection and spray application in near real-time.

Motivation & Objective

• To evaluate the feasibility of using YOLOv3 for detecting volunteer cotton plants in corn fields using RGB UAV imagery.
• To assess the performance of YOLOv3 across three image resolution scales (320×320, 416×416, 512×512) in terms of detection accuracy.
• To determine whether the model can support near real-time remote sensing applications for agricultural pest management.
• To investigate the impact of image scale on key metrics: average precision (AP), mean average precision (mAP), and F1-score.
• To support the development of automated systems for detecting and targeting volunteer cotton plants to prevent boll weevil reinfestation.

Proposed method

• Acquired three-band RGB aerial images using an unmanned aircraft system (UAS) over corn fields containing volunteer cotton plants.
• Trained a YOLOv3 object detection model on annotated aerial images to identify volunteer cotton plants at the pin-head square stage (5–6 leaf stage).
• Evaluated model performance at three fixed input image sizes: S1 (320×320), S2 (416×416), and S3 (512×512 pixels).
• Measured detection performance using standard metrics: average precision (AP), mean average precision (mAP), and F1-score at 95% confidence level.
• Performed statistical analysis (p-values) to compare model performance across the three scales.
• Utilized a computer vision-based remotely piloted aerial application system (RPAAS) framework for potential real-time deployment.

Experimental results

Research questions

• RQ1 Can YOLOv3 effectively detect volunteer cotton plants in corn fields using only RGB UAV imagery?
• RQ2 How does the input image resolution (320×320, 416×416, 512×512) affect the detection performance of YOLOv3 in terms of mAP, AP, and F1-score?
• RQ3 Is there a statistically significant difference in detection performance between the three image scales?
• RQ4 Can the trained YOLOv3 model support real-time detection and spray application in agricultural remote sensing systems?
• RQ5 Does the model maintain consistent mAP across different scales, indicating scalability for operational use?

Key findings

• No significant differences were found in mean average precision (mAP) across the three image scales (S1, S2, S3), indicating consistent model robustness.
• A statistically significant difference was observed in average precision (AP) between S1 and S3 (p = 0.04) and between S2 and S3 (p = 0.02).
• A significant difference was also found in F1-score between S2 and S3 (p = 0.02), suggesting better balance between precision and recall at S3.
• The lack of significant mAP differences across scales supports the use of the trained YOLOv3 model in near real-time applications on RPAAS platforms.
• The model demonstrates strong potential for deployment in automated, UAV-based systems for detecting and targeting volunteer cotton plants.
• The results confirm that YOLOv3 can be effectively used for early detection of volunteer cotton plants in corn fields using only RGB imagery.