Authors

Abstract

Soybean emergence uniformity is critical for yield formation, and the early acquisition of this information provides valuable guidance for field management. However, regional variations in cultivation patterns restrict most existing detection models to a single pattern, thereby limiting their practical applicability. To overcome this limitation, this study developed a universal detection system for assessing soybean emergence uniformity across diverse cultivation patterns. The system employs unmanned aerial vehicles (UAVs) to acquire field images, after which the YOLOv12 model is used to detect seedlings and extract their center coordinates. A two-stage clustering algorithm (Elliptical DBSCAN + K-means) is applied to classify seedlings into rows, and plant spacing is calculated by integrating Euclidean distance with ground sampling distance (GSD). Emergence uniformity is subsequently evaluated using the ISO-standard Multiple and Miss Indices. Field validation across ridge-based double-row, triple-row, and quadruple-row cultivation patterns yielded coefficients of determination (R2) of 0.9919, 0.9887, and 0.9924, respectively, with no significant differences compared to manual measurements (all p > 0.05). The system achieved plant-spacing detection accuracies of 96–97% during the low-occlusion VE (Vegetative Emergence) and VC (Vegetative Cotyledon) stages; however, accuracy for the quadruple-row pattern decreased to 79% at the V1 (Vegetative 1) stage due to severe leaf occlusion. This study presents the first detection framework applicable across multiple soybean cultivation patterns, providing a high-accuracy and reliable tool to support informed field management decisions.

Abstract in Chinese

大豆出苗均匀性是影响产量的重要因素，早期获取大豆出苗均匀性信息，可为田间管理提供指导。然而，不同地区栽培模式不同，大部分现有模型受限于一种栽培模式，存在严重局限性。为构建一种不同栽培模式下通用的大豆出苗均匀性检测系统，本研究提出一种方法，利用无人机采集大豆幼苗图像，基于YOLOv12模型检测幼苗并提取中心坐标；通过椭圆形DBSCAN和K-means两阶段聚类算法，将不同栽培模式下同一行大豆幼苗进行归类；结合欧几里得距离与地面采样分辨率(GSD)计算株距，并采用ISO标准的多重指数和缺失指数对大豆出苗均匀性进行评估。田间验证结果表明，本研究方法在垄上双行、垄上三行、垄上四行栽培模式下的株距检测决定系数R2分别为0.9919、0.9887、0.9924；均匀性指数与人工测量值无显著差异（卡方验证P值均＞0.05）；系统在低遮挡的出苗期、子叶期表现优异，株距数量检测准确率高达96%-97%，但在第一复叶时期叶面遮挡严重，垄上四行栽培模式的准确率降至79%。本研究提出的大豆出苗均匀性检测方法首次实现了在不同栽培模式下使用，且具有较高的准确率和可靠性，可以作为田间管理决策的依据。