基于改进YOLO的玉米幼苗株数获取方法

为快速准确获取玉米幼苗株数、评估播种质量、进行查缺补苗等管理，对YOLO算法进行改进，提出了一种基于特征增强机制的幼苗获取检测模型（FE-YOLO），实现了对玉米幼苗株数的快速获取。该方法根据玉米幼苗目标尺寸和空间纹理特征，构建了基于动态激活的轻量特征提取网络，融合了多感受野和空间注意力机制。实验表明：FE-YOLO模型增强了幼苗空间特征、降低了网络复杂度，使模型的mAP和召回率分别达到87.22%和91.54%，每秒浮点运算次数和检测推理时间仅为YOLO v3的7.91%和33.76%。FE-YOLO能够实现无人机正射影像的玉米幼苗株数获取和种植密度估算，该模型复杂度低、识别精度高，能够为玉米苗期管理提供技术支持。

Detection Method of Maize Seedlings Number Based on Improved YOLO

The number of maize seedlings is the essential information for sowing quality assessment. It is important to obtain the number of maize seedlings quickly and precisely for investigation and filling the gaps with seedlings. To improve the real time and precision of the acquisition of maize seedling number, the YOLO model (FE-YOLO) was improved, and the detection and acquisition of maize seedling number were realized. Firstly, dynamic ReLU was used to improve the bottleneck layer of MobileNet and the feature extraction performance of MobileNet was increased. Then, according to the target size and spatial texture characteristics of maize seedlings, the multi-receptive field fusion and spatial attention mechanism were used to enhance the feature expression. The experimental results showed that the FE-YOLO model enhanced the spatial texture characteristics of the seedlings, reduced the complexity of the model, made the mAP and recall rates reach 87.22% and 91.54%, respectively, and the floating-point operations per second and detection consumption time were only 7.91% and 33.76% of YOLO v3. FE-YOLO can detect the maize seedlings in the UAV orthoimage, and then Equation (13) was used to estimate the planting density. FE-YOLO had low complexity and high recognition accuracy, which can provide support for maize seedling management.