This paper presents a wheat seed detection and counting method based on an improved YOLOv8 model, aiming to address the challenges of seed accumulation, adhesion, and occlusion in complex agricultural environments. The proposed YOLOv8-HD model incorporates shared convolutional layers to reduce parameter count and enhance runtime speed, and integrates a Vision Transformer with a Deformable Attention mechanism into the C21 module of the backbone network to improve feature extraction capabilities. The results show that YOLOv8-HD achieves an average detection accuracy (mAP) of 77.6% in stacked scenes with impurities and 99.3% across all scenes, outperforming the original YOLOv8 model. The model's memory size is 6.35 MB, and its inference time is 2.86 ms on a GPU, making it suitable for real-time counting on embedded platforms. Extensive experiments demonstrate that YOLOv8-HD outperforms other mainstream networks in terms of mAP, speed, and model size, providing robust technical support for the development of seed counting instruments.This paper presents a wheat seed detection and counting method based on an improved YOLOv8 model, aiming to address the challenges of seed accumulation, adhesion, and occlusion in complex agricultural environments. The proposed YOLOv8-HD model incorporates shared convolutional layers to reduce parameter count and enhance runtime speed, and integrates a Vision Transformer with a Deformable Attention mechanism into the C21 module of the backbone network to improve feature extraction capabilities. The results show that YOLOv8-HD achieves an average detection accuracy (mAP) of 77.6% in stacked scenes with impurities and 99.3% across all scenes, outperforming the original YOLOv8 model. The model's memory size is 6.35 MB, and its inference time is 2.86 ms on a GPU, making it suitable for real-time counting on embedded platforms. Extensive experiments demonstrate that YOLOv8-HD outperforms other mainstream networks in terms of mAP, speed, and model size, providing robust technical support for the development of seed counting instruments.