Maize plant detection using UAV-based RGB imaging and YOLOv5

Abstract

In recent years, computer vision (CV) has made enormous progress and is providing great possibilities in analyzing images for object detection, especially with the application of machine learning (ML). Unmanned Aerial Vehicle (UAV) based high-resolution images allow to apply CV and ML methods for the detection of plants or their organs of interest. Thus, this study presents a practical workflow based on the You Only Look Once version 5 (YOLOv5) and UAV images to detect maize plants for counting their numbers in contrasting development stages, including the application of a semi-auto-labeling method based on the Segment Anything Model (SAM) to reduce the burden of labeling. Results showed that the trained model achieved a mean average precision (mAP@0.5) of 0.828 and 0.863 for the 3-leaf stage and 7-leaf stage, respectively. YOLOv5 achieved the best performance under the conditions of overgrown weeds, leaf occlusion, and blurry images, suggesting that YOLOv5 plays a practical role in obtaining excellent performance under realistic field conditions. Furthermore, introducing image-rotation augmentation and low noise weight enhanced model accuracy, with an increase of 0.024 and 0.016 mAP@0.5, respectively, compared to the original model of the 3-leaf stage. This work provides a practical reference for applying lightweight ML and deep learning methods to UAV images for automated object detection and characterization of plant growth under realistic environments.

Keywords: UAV remote sensing; YOLOv5 application; crop scouting; deep learning; plant detection model.

Figure 1

Study site and the workflow of maize plant detection.

Figure 2

Image annotations: Generated by SAM (A) and adjusted manually by LabelImg (B). The red rectangle represents maize seedlings, the blue circle represents erroneous identification by SAM, and the orange circle represents missed maize seedlings by SAM.

Figure 3

Acquired UAV Images: 3-leaf stage (A), 3-leaf stage with low noise (B), and 7-leaf stage (C).

Figure 4

Model evaluation metrics: the object loss (A), the bounding box loss (B), the Precision (C), the Recall (D), the mAP@0.5 (E), and the mAP@0.5:0.95 (F). The thin line represents raw data, while the thick curve represents the results after local weighted regression scatter smoothing.

Figure 5

Comparison of the final performance of the models: Comparison of accuracy, including mAP, Precision, and Recall (A) and comparison of loss (B).

Figure 6

Practical performance of the 3-leaf stage model: origin model (A), origin model in overgrown weeds conditions (B); 90-degree rotation model in normal conditions (C), origin model with a pre-training weight of 90-degree rotation model (D) low noise model in normal conditions (E), and origin model with a pre-training weight of low noise model (F).

Figure 7

Confusion Matrix: 3l_origin (A), 3l_90d (B), 3l_origin_90d (C), 3l_nonoise (D), 3l_origin_nonoise (E), 7l_origin (F), and 7l_90d (G). 7l_origin_90d (H). The rows represent the prediction results, the columns represent the ground truth, and the number in the grid is the number of objects.

Figure 8

Practical performance of the 7-leaf stage model: origin model (A), 90-degree rotation model (B), and origin model with a pre-training weight of 90-degree rotation model (C). The red rectangle represents the models’ predicted results, while the orange circle represents the maize plants that were missed by the model.