SEGMENTATION AND PREDICTION OF TIME PATTERNS OF LOW LEVEL TRAILS Russian patent published in 2022 - IPC G01M3/02 G06N3/08 G06T7/00 

FIELD: gas industry.

SUBSTANCE: invention relates to a system and a method for detection of gas trail based on image data. In the method, data of images related to gas trail is received, wherein image data includes a set of image frames; trail analysis data is determined using received image data and the first prediction model trained to receive image data, and trail analysis data related to gas trail is generated in response to reception. In this case, trail analysis data includes a mask of predicted trail segmentation, related to gas trail and predicted relatively to a sequence of image frames included in image data. At the same time, the specified mask of predicted trail segmentation includes pixel-by-pixel segmentation establishing a border of gas trail, predicted by means of the first prediction model, which is superimposed on pixel-by-pixel representation of a reference segmentation mask including annotation data related to the sequence of image frames and provided by means of training the first prediction model by a user. At least one of the following is determined: assessment of a leakage size, assessment of a leakage source, and assessment of a volumetric leakage speed for gas trail. Assessment of the leakage size, assessment of the leakage source, assessment of the volumetric leakage speed, and trail analysis data are provided, wherein trail analysis data is represented in the form of superimposition on image data. Trail analysis data superimposed on image data is transmitted. In this case, at least one of reception, determination, provision, and transmission is performed by at least one data processing processor, which is part of at least one computer system.

EFFECT: increase in the accuracy of detection of leakage for gas trail based on image data.

18 cl, 15 dwg, 1 tbl