基于SegFormer的传统水表读数区域分割方法

打开文本图片集

中图分类号：TP391.41 文献标志码：A 文章编号：2095-2945（2025）29-0058-04

Abstract：During theupgradeofsmartwatersystemsandthedigitaltransformationofoldwater meters，achievinghighly robustreadingareasegmentationisakeypretreatmentstepforautomatedmeterreading.Inviewoftheinterferencefactorsuch asthecomplexinstalationenvironmentoftewatermeter，thedialisrfletedbyglassandblockedbywaterstains，thispaper proposesalightweightsegmentationmodelbasedonSegFormer，focusingonpixel-levelpositioningofthewatermeterreading area.Byoptimizingmodelparametersbasedonwatermeterimagefeatures，thismethodsignificantlyimprovessegmentation accuracyandedgecontinuityundercomplexworkingconditions.Experimentsonaself-builtwatermeterimagedatasetshowthat this method achieves a segmentation accuracy of 98.60% on the test set，and compresses the model parameter quantity to 14.0M Thismethodcanprovidehigh-confidenceregionalinputforsubsequentwater meterreadingidentificationmodules，andprovides anefcientandreliablesegmentationpretreatmentschemeforthedigitaltransformationoftraditionalwatermeters.Inaddition， this technical framework can be extended to pan-industrial testing scenarios such as gas meters and pressure gauges.

Keywords：smartwater;SegFormer;digitizationoftraditionalwatermeter;readingareasegmentation;pixel-levelclasification

随着智慧水务的快速发展，智能远传水表的功能越来越完善。（剩余5560字）