A Method for Marking Abnormal Areas of Industrial Processes Based on Weighted Median Filtering

Abstract

The present invention proposes a method for marking abnormal areas of industrial processes based on weighted median filtering, which specifically includes: acquiring RGB format video image sequences, converting the RGB format video image sequences into grayscale video images; sparsely and roughly marking the video image sequences; sparsely and roughly Marker propagation; optimize the rectangular marker area of ​​all video images; obtain the final label data; by only manually marking the start frame and end frame of a video sequence, it is possible to mark the semi-melting conditions in the entire video, Get labeled data, enabling rapid acquisition of large volumes of data with semi-melted labels. In the diagnosis of semi-melting conditions, a large amount of labeled data improves the training accuracy of the model, thereby improving the accuracy of diagnosis. Promote the application of diagnostic technology in industrial sites, save manpower and material resources, and reduce potential safety hazards.

Description

technical field [0001] The invention belongs to the technical field of fault diagnosis, in particular to a method for marking abnormal regions of industrial processes based on weighted median filtering. Background technique [0002] Fused magnesia, also known as fused magnesia (MgO), has the characteristics of high melting point, strong structure, strong insulation, and good oxidation resistance. It is an important refractory material in many industries such as metallurgy, chemical industry, electrical equipment, and aerospace. At present, its production process is mainly to use a three-phase AC electric fused magnesia furnace to heat and melt magnesite ore to obtain MgO melt, and then obtain high-purity fused magnesia through cooling and crystallization. The abnormal working condition of the semi-melting of the fused magnesium furnace is caused by the condensation of the high-temperature magnesium oxide melt on the inner wall of the furnace shell, and the phenomenon that th...

AI Technical Summary

Problems solved by technology

However, such methods need to use video images as input data and label data to train the network, and then adjust network parameters through data that contains both input and label data, so synchronously sampled working condition video data and a large number of working condition label data are required.

Although the video image data can be obtained in real time at a high rate, the label data of the corresponding period of time is often sparsely marked and difficult to obtain

At present, there is no method to quickly generate labeled data. The existing method is to manually judge whether a semi-melting condition occurs, and then artificially label each frame of the training set video sequence

Manually making label data not only requires a lot of manpower and high cost, but also takes a long time due to the low efficiency of manual labeling

Images