基于改进YOLOv4算法的工程实验室信息化管理系统构建与应用

Construction and application of engineering laboratory information management system based on improved YOLOv4 algorithm

融合物联网技术、图像检测技术和深度学习算法,提出一种基于改进YOLOv4算法的工程检测实验室信息管理系统,通过WIFI+NB-IoT实现数据传递。为了弥补YOLOv4算法在尺度分布不均匀时精度降低的问题,提出利用IK-means++算法,引入ECA注意力模块和阶梯状特征融合网络结构对算法进行改进,算法改进后浮点运算数量、模型参数量分别降低了25.1%和43.1%,FPS和mAP分别提高6.8帧/s和3.65%,改进后算法不仅收敛速度更快,而且在不同光线环境下的设备检测准确率均高于改进前。将系统应用到工程实验室检测中,设备和环境各参数检测结果与标准仪器检测结果误差控制在±5%以内。

Integrating Internet of Things technology,image detection technology and deep learning algorithm,an information management system for engineering testing laboratories based on improved YOLOv4 algorithm was proposed,and the data transmission was realized through WIFI+NB-IoT.In order to make up for the problem of the reduced accuracy of the YOLOv4 algorithm when the scale distribution was uneven,the IK-means++algorithm was proposed,and the algorithm was improved by introducing the ECA attention module and the stepped feature fusion network structure,and the number of floating-point operations and model parameters were reduced by 25.1%and 43.1%,respectively,and the FPS and mAP were increased by 6.8 frames/s and 3.65%,respectively.The improved algorithm not only converged faster,but also had a higher detection accuracy than before the improvement in different light environments.The system was applied to the engineering laboratory testing,and the error between the test results of various parameters of the equipment and environment and the test results of standard instruments was controlled within±5%.