基于改进YOLOv7的混凝土工厂障碍物识别  

Obstacle Recognition in Concrete Factories Based on Improved YOLOv7

作　　者：张杰 杨俊雅 李冬冬 程雪斌 Zhang Jie;Yang Junya;Li Dongdong

机构地区：[1]中交二航局第一工程有限公司,湖北省武汉市430040 [2]中交第二航务工程局有限公司

出　　处：《工程机械》2025年第4期32-41,I0002,共11页Construction Machinery and Equipment

摘　　要：随着人工智能技术的快速发展,目标检测算法在工业领域中的应用越来越广泛。混凝土工厂环境复杂,障碍物种类多样,对障碍物的准确识别对于保障生产安全至关重要。提出一种基于改进YOLOv7的混凝土工厂障碍物识别算法,通过引入SPD-Conv模块与GAM-Attention,在不降低模型检测性能的前提下,优化算法对混凝土障碍物的识别精度。试验结果表明,与原始YOLOv7模型相比,在6个不同障碍物类别上准确率、召回率、综合性的评估指标F1和评估模型性能的核心指标mAP@0.5分别提升了1.0%~10.4%、2.8%~11.3%、0.5%~10.5%、3.5%~8.3%,表明该算法具有更高的识别精度。Improved-YOLOv7的mAP@0.5和mAP@0.5:0.95分别为0.908和0.710,参数量为36.91 MB,Flops为135.2 G,ADT为24.9 ms,相较于Ca s ca de R-CNN、Fa s te r R-CNN、Re tina Ne t、S S D、YOLOX模型,Improved-YOLOv7模型具有更高的识别精度和性能,对于算力需求最小,能够有效识别混凝土工厂中的各类障碍物,为实际生产提供有力支持。With the rapid development of artificial intelligence technology,object detection algorithms are more and more widely used in the industrial field.Concrete factories have complex environments with various types of obstacles,so accurate recognition of obstacles is crucial for ensuring production safety.A concrete factory obstacle recognition algorithm based on the improved YOLOv7 is proposed,which optimizes the recognition accuracy of concrete obstacles by introducing the SPDConv module and GAM-Attention without reducing the model detection performance.The test results show that,compared with the original YOLOv7 model,the accuracy,recall and comprehensive evaluation indicator F_(1)and core indicator mAP@0.5 for evaluating the model performance on six different obstacle categories are improved by 1.0%to 10.4%,2.8%to 11.3%,0.5%to 10.5%and 3.5%to 8.3%,respectively,indicating that the algorithm has higher recognition accuracy.The mAP@0.5 and mAP@0.5:0.95 of Improved-YOLOv7 are 0.908 and 0.710,respectively,with params of 36.91 MB,Flops of 135.2 G and ADT of 24.9 ms,and compared with Cascade RCNN,Faster R-CNN,RetinaNet,SSD and YOLOX models,the Improved-YOLOv7 model has higher recognition accuracy and performance and requires minimal computing power,which can effectively recognize various obstacles in concrete factories and provide strong support for practical production.

关 键 词：目标检测 YOLOv7 混凝土工厂 障碍物识别 注意力机制 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]