基于改进YOLOv8-seg的工业零件实例分割算法研究

Research on Industrial Parts Instance Segmentation Method Based on Improved YOLOv8-seg

摘要:

随着工业自动化和智能制造技术的快速发展，在复杂工业场景之下工业零件的精准实例分割对装配质检尤为重要；针对工业场景下因光照变化大、相机视角变化大以及局部遮挡因素导致的零件实例分割精准性与鲁棒性不足的问题，提出一种基于改进YOLOv8-seg的工业零件实例分割算法模型；在主干网络中引入LIR-CAGM模块，降低光照变化对特征提取的影响，增强模型对结构稳定信息的表征能力;在颈部网络中采用DCNv2-C2f模块替换原有的C2f模块，提高模型对视角变化与局部遮挡场景的适应性；将原本的损失函数CIoU替换成SA-CIoU，增强目标定位的优化稳定性。经过实验检测，改进后的模型在测试数据集上的mAP:0.5以及mAP0.5:0.95分别达到了92.1%和61.8%的分割效果，相较于原YOLOv8-seg网络分别提高了4.5%和4.4%，验证了该模型的有效性。

Abstract:

With the rapid development of industrial automation and intelligent manufacturing technologies, precise instance segmentation of industrial parts under complex industrial scenarios is crucial for assembly quality inspection; to address the insufficient accuracy and robustness of part instance segmentation caused by large illumination variations, significant camera viewpoint changes, and partial occlusions, an industrial part instance segmentation algorithm based on improved YOLOv8-seg is proposed; the LIR-CAGM module is introduced in the backbone network to reduce the impact of illumination variations on feature extraction and enhance the representation of structurally stable information; the DCNv2-C2f module is adopted in the neck network to replace the original C2f module, improving the model’s adaptability to viewpoint changes and occluded scenes; the original CIoU loss is replaced by SA-CIoU to enhance the optimization stability of target localization. Experimental results show that the improved model achieves mAP@0.5 and mAP@0.5:0.95 of 92.1% and 61.8% on the test dataset, respectively, representing improvements of 4.5% and 4.4% compared to the original YOLOv8-seg network, validating the effectiveness of the proposed model.