Anomaly Detection Method of Urban Drainage Pipeline Video Based on Multi-Instance Learning

Abstract

The invention relates to the technical field of pipeline anomaly detection, and discloses a multi-instance learning-based video anomaly detection method for urban drainage pipelines. img ∈ R B×K×H×W×C with X diff ∈ R B×K×H×W×C ;According to the input X of the method img ∈ R B×K×H×W×C with X diff ∈ R B×K×H×W×C , calculate the output of the method and according to the output of the method and the input of the method, calculate the gap L between the output of the method and the true value of the input data; according to the gap L, calculate the parameter θ of the update method; the optimization of the iterations required by the given method The number of times is T, and if the number of iterations reaches T times, the method optimization process ends. Automatically reviewing whether there are abnormalities in the video of the drainage pipe by the computer can save a lot of labor costs, and the time required to process a single video is shorter, which can save the time of reviewing the video of the drainage pipe. After a large amount of data training, the processing result will be more stable. , which improves the detection accuracy of video anomaly detection methods for urban drainage pipes.

Description

technical field [0001] The invention relates to the technical field of pipeline anomaly detection, in particular to a video anomaly detection method for urban drainage pipelines based on multi-instance learning. Background technique [0002] Urban drainage pipes are important facilities to ensure the normal life of urban residents. With the gradual expansion of the city area, more and more sewage, rainwater, waste or residues need to be discharged in the city, and the length of urban drainage pipes is also increasing. With the rapid growth of the urban area, the anomaly detection and maintenance of the drainage pipeline has brought a huge workload. [0003] In the prior art, the anomaly detection task of the drainage pipe mainly relies on manually reviewing and collecting the video inside the drainage pipe. However, due to the expansion of the urban area, the number of drainage pipe videos that need to be manually reviewed every day increases rapidly, which requires more ma...

AI Technical Summary

Problems solved by technology

Among them, foreign object insertion, ups and downs and wrong mouths, these types of abnormalities are easily missed by manual review, so they bring certain deviations to the abnormal detection results of drainage pipes and affect the abnormal detection accuracy of drainage pipe video

Images