Assessing the impact of soil and field conditions on cotton crop emergence using UAV-based imagery

Abstract

Crop seeding rate is one of the crucial factors that affect crop production. However, acquiring adequate crop data in multiple growing environments is time-consuming and challenging in large fields. This study aimed to develop and evaluate an efficient method using an unmanned aerial vehicle (UAV) imaging system and deep learning to assess cotton emergence spacing uniformity at different seeding rates. The study was conducted on a 3.27-hectare research field planted with two cotton cultivars at five seeding rates (56 k, 74 k, 91 k, 108 k, and 123 k seeds ha⁻¹), with each treatment containing four rows with three replicates in a random block design. A UAV imaging system collected RGB images at 10 m and 15 m flight height above the ground level at two and six weeks after planting. Orthomosaic images from the two days were segmented into small blocks that were processed using the object detection algorithm YOLOv7 to identify cotton plants. Hough transform and polynomial regression were used to identify each cotton row and remove weeds. The number of plants in each 5-m row segment (i.e., stand count) was calculated to correlate with soil electrical conductivity (ECₐ) and field elevation. Results show that the research could detect cotton plants with the mean average precision of 96.9 % at the 50 % intersection over the union threshold (mAP@50) for the two-week dataset and 92.7 % mAP@50 for the six-week dataset. The results also show that plant uniformity was closely correlated with field elevation and ECₐ, with an average R² of 0.62 using the Random Forest model. The coefficient of variation was used to evaluate the spacing uniformity of each seeding rate and demonstrated that the seed rates of 108 k and 123 k seeds ha⁻¹ tended to exhibit better spacing uniformity than others under various environmental conditions. This study provides valuable insights by developing a pipeline for early-stage cotton stand count using high-resolution remote sensing techniques to evaluate the uniformity of different seeding rates for cotton, ultimately improving the efficiency of crop management.

Plant phenotyping relevance

UAV画像と深層学習・画像処理によりワタ個体を検出し、出芽数と株間均一性を定量化するパイプラインの開発・評価が中心であるため、植物表現型計測手法として採用する。

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.