Urban underground pipeline is one of the important infrastructures in the city,and the timely detection of pipeline defects plays a more important role in the development of the city.Aiming at the current pipeline defect detection model with large parameter number and poor real-time performance,an improved FCOS method of the defect detection is proposed for urban underground pipelines.Firstly,the lightweight MobileOne network is introduced,and the model size is reduced by converting the multi-branch network into a single-branch network through structural reparameterization.Then,the joint branch of classification and IoU is introduced to make that the model's training is consistent with inference process,and the balancing factor is utilized to optimize the QFL,which improves the classification prediction effect of the model.The experimental results show that the improved FCOS model improves the average accuracy by 1.83%compared with the baseline model,the detection speed FPS reaches 48.6,and the number of model parameters decreases by 17.85 M,which effectively improves the performance of defects detection for urban underground pipelines,and it also has certain advantages compared with other excellent target detection algorithms.
urban underground pipelinedefect detectionFCOS algorithmreparameterizationjoint branchQFL loss function
维普
万方数据
