【摘要】 针对铸件打磨机械臂中实现铸件浇冒口残留区域图像分割的需求，制作铸件浇冒口图像数据集，完成了YOLOv8模型改进工作，将模型主干替换为ShuffleNet,提出了Shuffle-Block-LKA和C2fLKA两种模块用于提高特征提取能力，并向YOLOv8n-seg-ShuffleNet模型加入了BiFPN特征融合结构，以增强模型的特征融合能力。经实验验证，改进后的模型表现出良好的检测效果和较快的检测速度，在ONNXRuntime推理引擎的优化部署下能够满足实时性要求。更多还原

【Abstract】 Casting is one of the important molding processes, yet the traditional grinding process is low in automation. In recent years, automatic grinding robots have gradually developed. To address the need for image segmentation of residual areas of casting risers in casting grinding robotic arms, a dataset of casting riser images was created, and improvements were made to the YOLOv8 model. The backbone of the model was replaced with ShuffleNet, and two modules, Shuffle-Block-LKA and C2fLKA, were proposed to enhance feature extraction capabilities. Additionally, the BiFPN feature fusion structure was incorporated into the YOLOv8n-seg-ShuffleNet model to strengthen its feature fusion ability. Experimental results demonstrate that the improved model exhibits good detection performance and fast detection speed, meeting real-time requirements under the optimized deployment of the ONNX Runtime inference engine.更多还原