With the deepening of industrial automation, a large number of edge intelligent devices are deployed in industrial meter detection. In view of the limited computing and storage capacity of these embedded devices, we propose a lightweight meter detection method. Our proposed method is based on the widely used Yolov5, the depthwise separable convolution and squeeze and excitation channel attention module are used to simplify the backbone and head of the network, and further prune the filters of convolution layers via geometric median. Finally, model parameters and floating-point operations are reduced to 0.250M and 0.687G on the premise of ensuring the effect of the meter detection.