SYSTEM AND METHOD FOR IDENTIFICATION OF FEATURES IN IMAGES OBTAINED BY DRONE INSPECTIONS

    公开(公告)号:US20240212318A1

    公开(公告)日:2024-06-27

    申请号:US18069880

    申请日:2022-12-21

    CPC classification number: G06V10/764 G06V20/17

    Abstract: A method is described for automatic creation of image classifiers using unannotated data. The method may include receiving a CLIP model configured to generate images from text; using the CLIP model to generate images of pump jacks and images of oil pools; providing the training pairs comprising the images of pump jacks and the text “pump jack” and the images of oil pools and the text “oil pool” to a classification model; and training the classification model using the training pairs for contrastive learning to generate a refined model. The refined model may be used to classify images provided by drones to identify possible oil pools. The method is executed by a computer system.

Patent Agency Ranking