MaizeStandCounting (MaSC): Automated and Accurate Maize Stand Counting from UAV Imagery Using Image Processing and Deep Learning

TL;DR

MaizeStandCounting (MaSC) tackles automated maize stand counting from low-cost UAV RGB imagery by combining two input pathways (mosaic patches and raw video frames) with a lightweight YOLOv9 detector trained on seedling growth stages. The pipeline integrates ExG-based color segmentation and Radon-derived row orientation with range/row detection to yield per-row counts, and it analyzes performance against manual ground truth. Raw-frame processing outperforms mosaic-based counting, achieving a strong $R^2$ of $0.906$ and demonstrating near real-time capability, while mosaic mode is more sensitive to stitching artifacts. The work offers a practical, scalable approach for high-throughput maize phenotyping in research and production, with public data and code to enable adoption and extension.

Abstract

Accurate maize stand counts are essential for crop management and research, informing yield prediction, planting density optimization, and early detection of germination issues. Manual counting is labor-intensive, slow, and error-prone, especially across large or variable fields. We present MaizeStandCounting (MaSC), a robust algorithm for automated maize seedling stand counting from RGB imagery captured by low-cost UAVs and processed on affordable hardware. MaSC operates in two modes: (1) mosaic images divided into patches, and (2) raw video frames aligned using homography matrices. Both modes use a lightweight YOLOv9 model trained to detect maize seedlings from V2-V10 growth stages. MaSC distinguishes maize from weeds and other vegetation, then performs row and range segmentation based on the spatial distribution of detections to produce precise row-wise stand counts. Evaluation against in-field manual counts from our 2024 summer nursery showed strong agreement with ground truth (R^2= 0.616 for mosaics, R^2 = 0.906 for raw frames). MaSC processed 83 full-resolution frames in 60.63 s, including inference and post-processing, highlighting its potential for real-time operation. These results demonstrate MaSC's effectiveness as a scalable, low-cost, and accurate tool for automated maize stand counting in both research and production environments.

Figures (11)

• Figure 1: Workflow of segmentation with color.
• Figure 2: Overview of the MaSC processing workflow. Green boxes represent the pipeline for pre-mosaicked image inputs. The blue box indicates the raw video input mode, which includes internal mosaicking. Orange boxes show the seedling detection process using YOLOv9, shared across both input modes.
• Figure 3: Training results of YOLOv9. The plots show the evolution of box loss, classification loss, and distribution focal loss during training (left), along with the corresponding performance metrics: precision, recall, mean average precision at IoU=0.5 (mAP@0.5), and mean average precision at IoU=0.5:0.95 (mAP@0.5:0.95) (right).
• Figure 4: Color-based segmentation under ideal field conditions. The field section contains only maize, with minimal weed interference and well-separated plants. Red lines denote Voronoi boundaries, and yellow arrows indicate multiple plants within a single cell.
• Figure 5: Close-up view of the 2020 dataset showing maize seedlings interspersed with weeds and grass.