基于轻量化改进的YOLOv7网络下仪表定位检测研究

Research on Instrument Positioning and Detection Based on Lightweight Improved YOLOv7 Network

为了高效准确地完成工业指针式仪表表盘的定位检测,研究了轻量化的YOLOv7目标检测网络算法。根据仪表图像中目标形状特征多样的特点,对网络卷积核进行了轻量化改进,使用深度可分离卷积代替普通卷积减少参数量,提高网络模型检测精度和检测速度,并与其它网络模型进行了性能指标对比。实验表明,轻量化改进后的YOLOv7网络mAP指标上提升了3.03%,参数量和模型规模下降了35.1%,准确率提高了4.61%,检测速度提升了21.3%,实现了高效率高精度的仪表检测。

In order to efficiently and accurately detect the dial of industrial pointer instrument,a lightweight YOLOv7 target detection network algorithm is studied.According to the various shape features of the target in the instrument image,the network convolution kernel is improved by using the depth separable convolution instead of the common convolution to reduce the number of parameters and improve the detection accuracy and speed of the network model.Experiments show that the mAP index of the lightweight and improved YOLOv7 network is increased by 3.03%,the number of parameters and model size are decreased by 35.1%,the accuracy is increased by 4.61%,and the detection speed is increased by 21.3%,realizing the high efficiency and high precision instrument detection.