Corrosion degree detection of power tower poles based on machine vision
In response to the problem of surface corrosion of power tower poles in different environments,an instance segmentation method based on YOLACT is introduced by using machine vision technology.ResNet is used as the backbone network to extract image features,and multi-loss functions and Fast NMS are used to improve the segmentation performance of the model.YOLACT is used to establish the baseline,training is conducted at different backbone networks and resolutions,the ResNet101-700×700 model with better average accuracy is selected.The RGB color characteristics of corrosion are deduced from the acquired images and the corrosion degree is divided.A boundary box strategy is employed for corrosion detection on the towers.The concept of linear projection is introduced to better localize the corroded areas.Validation results on the test dataset achieve an average precision of 97.3%and an average recall of 94.1%.
NETL
NSTL
万方数据
