Deep learning-based wheat ear recognition and counting

In response to the high missed detection rate of wheat ears and significant background interference in complex farmland scenarios, this study proposes an improved solution based on multi-scale feature fusion and a mixed attention mechanism: YOLOv11s-ASF-DynamicScalSeq-SE. By constructing a multi-scale feature pyramid network to achieve cross-layer feature interaction, the model's sensitivity to detecting wheat ears of different sizes is effectively enhanced; at the same time, DynamicScalSeq, ASF, and SE are integrated, forming a network architecture with hierarchical feature calibration capability. Experimental results show that the improved model achieved an accuracy of 91.9% and a recall rate of 89% in the construction of the wheat ear dataset, which are higher than the baseline model by 0.2 and 0.1 percentage points, respectively, and its robustness in complex field scenarios is significantly enhanced.