基于AI目标检测在工程管理图像数字化的应用实践

Application Practice of AI-based Target Detection in the Digitalization of Engineering Management Images

为强化数据驱动的工程管理,推进工程建设数字化、智慧化以及工程实施和管理的标准化,需借助大数据AI技术实现工程施工过程中的图像目标检测,以提升工程质量和效率。本文探讨不同AI目标检测算法在工程质检场景的应用,通过比对两阶段检测算法Faster RCNN和Cascade RCNN、单阶段检测算法yolo系列在空开熔丝设备型号规格检测的实验效果,yolov5在本数据集上的检测速率(Fps)优于Faster RCNN、Cascade RCNN、PPyolo,而yolov5的均值平均精度mAP(0.981)与Faster RCNN(0.961)、Cascade RCNN(0.978)、PPyolo(0.986)表现相当,yolov5兼备了检测速率和检测精度。

In order to strengthen data-driven engineering management, promote digitalization and intelligence, as well as standardization of engineering implementation and management, AI technology is needed to realize target detection of image in the process of engineering construction, so as to improve the quality and efficiency of engineering. This paper discusses the application of different AI target detection algorithms in engineering quality inspection scenarios. By comparing the experimental effects of two-stage detection algorithms Faster RCNN, Cascade RCNN and single-stage detection algorithm YOLO series in the model and specification detection of Air switch and fuse equipment, the detection rate(Fps) of Yolo V5 in this dataset is better than Faster RCNN, Cascade RCNN and PPyolo. The mean accuracy m AP(0.981) of YOLOV5 is equivalent to that of Faster RCNN(0.961), Cascade RCNN(0.978) and PPyolo(0.986). Yolov5 has both detection rate and detection accuracy.