Authors

Abstract

Accurately counting the number of corn seedlings is a key step in assessing the growth status of corn. Due to the fact that the emergence of corn in the field is easily covered by straw and weeds, it is difficult to ensure the accuracy of emergence by relying on artificial features. In response to these problems, a method for detecting and counting corn seedlings in unmanned aerial vehicle remote sensing images based on the MEI-YOLOv11 model is proposed. The MANet module is introduced to enhance the feature extraction ability in complex environments. The EUCB module is adopted to optimize the recognition of small target details. The Inner-SIoU loss function is utilized to improve the convergence efficiency. Combined with data augmentation, a full weather dataset is constructed to enhance robustness. The results show that the mAP0.5, P and R of this model reach 97.0%, 94.2% and 95.7% respectively, which are 2.8, 2.7 and 2.4 percentage points higher than those of the original model respectively. The number of parameters and the reasoning time only increased by 1.28M and 0.4ms respectively. This method can effectively identify corn seedlings in complex backgrounds and meet the requirements of real-time detection at the same time. Its performance is significantly superior to other detection models. The counting accuracy rate was above 90% in multiple weather conditions, with the highest accuracy rate in sunny conditions, reaching 91.23% (RMSE=4.5044, R²=0.8508). The results can provide technical support for the accurate detection and counting of corn seedlings in multiple weather conditions.

Abstract in Chinese

准确统计玉米苗数是评估玉米生长状况的关键环节。由于田间玉米的出苗容易被秸秆和杂草覆盖，依靠人工特征难以保证出苗的准确性。针对这些问题，提出了一种基于MEI-YOLOv11模型的无人机遥感影像玉米幼苗检测与计数方法。引入MANet模块增强复杂环境的特征提取能力，采用EUCB模块优化小目标细节识别，利用Inner-SIoU损失函数提高收敛效率，结合数据增强构建全天气数据集，提高鲁棒性。结果表明，该模型的mAP0.5、P和R分别达到97.0%、94.2%和95.7%，比原模型分别提高了2.8、2.7和2.4个百分点。参数个数和推理时间分别只增加了1.28M和0.4ms。该方法能够有效识别复杂背景下的玉米幼苗，同时满足实时检测的要求，其性能明显优于其他检测模型。多天气环境下计数准确率均在90%以上，晴天环境下准确率最高，达到91.23% (RMSE=4.5044, R²=0.8508)，其结果可为多天气玉米幼苗准确检测计数提供技术支持。