摘要
针对实际的电石业务场景,基于YOLO系列模型设计了电石检测分割的方法,并开发了相应的系统。首先,通过采集大量的图像样本构建了较大规模的数据集并进行标注;为了节省算力和提高准确率,将检测和分割分为两个模块,模型分别为YOLOv5-l和YOLOv5-seg-m。在实验中,对模型的内部结构进行了一系列的改进,提高了算法的运算速度;并在此基础上对模型进行了参数量化,进一步节省了带宽、降低了模型的存储和运算所需的空间。相比于MaskR-CNN算法,在分割效果接近的情况下,速度得到了极大的提升。很好地完成了检测和分割电石的任务,为后续的生产自动化和无人化打下了基础,填补了相关检测的空白。之后将会针对排除光照影响、提高分割精度和评判电石质量模型继续研究。
A method for detecting and segmenting calcium carbide using YOLO series models was designed and a corresponding system was developed for practical carbide industry scenarios.First,a large-scale dataset was constructed by collecting a large number of image samples and annotating them,and to save computing power and improve accuracy,detection and segmentation were separated into two modules using the YOLOv5-l and YOLOV5-seg-m models respectively.In experiments,the internal structure of the models was improved to increase the computational speed,and the model was parameterized to further save bandwidth,storage,and computation space.Compared to the Mask R-CNN algorithm,the speed was greatly improved with similar segmentation results.The task of detecting and segmenting calcium carbide was successfully completed,laying the foundation for subsequent production automation and unmanned operation.Further research will be conducted to eliminate the influence of lighting,improve segmentation accuracy,and evaluate carbide quality models.
作者
郝俊峰
李玉涛
来博文
Hao Junfeng;Li Yutao;Lai Bowen(Artificial Intelligence Studio,HBIS Digital Technology Co.,Ltd.,Shijiazhuang 050035,China;Technical Center Headquarter,HBIS Digital Technology Co.,Ltd.,Shijiazhuang 050035,China)
出处
《现代计算机》
2023年第16期1-7,14,共8页
Modern Computer
