Foreign matter data generation method and terminal

Abstract

The invention discloses a foreign matter data generation method. A power transmission line scene image and a preset foreign matter image are acquired; extracting a power transmission line in the power transmission line scene image to obtain a power transmission line position; determining a target power transmission line position from the power transmission line positions, and pasting the preset foreign matter image into the target power transmission line position to obtain initial foreign matter data; based on the initial foreign matter data, a preset image harmony neural network is used for harmony to obtain final foreign matter data, foreign matter data generation on various different backgrounds can be achieved, the number is not limited, and finally based on the initial foreign matter data, the preset image harmony neural network is used for harmony to obtain the final foreign matter data. Foreign matters and scenes in the finally obtained foreign matter data can be better fused, and the real effect of the foreign matter data is improved, so that the quantity of the foreign matter data in a specific power transmission line scene is effectively increased, training samples are enriched, and the problem of migration performance reduction caused by different training and testing scenes is avoided.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a method for generating foreign object data and a terminal. Background technique [0002] Target detection is one of the most important tasks in computer vision, among which, the target detection of foreign objects in power lines has important research value. Common foreign objects on power lines include kites, plastic bags and tree branches. The difficulties of the current detection of foreign objects on transmission lines are: (1) few foreign object data; (2) unbalanced types of foreign objects; (3) ever-changing forms of foreign objects, which further exacerbates the problem of insufficient data. [0003] At present, most foreign object detection methods based on deep learning have the problem of lack of data. In order to solve the data shortage problem, they collect foreign object data of power lines from all different places, use it to train the network, an...

AI Technical Summary

Problems solved by technology

The difficulties of the current detection of foreign objects on transmission lines are: (1) few foreign object data; (2) unbalanced types of foreign objects; (3) ever-changing forms of foreign objects, which further exacerbates the problem of insufficient data

[0003] At present, most foreign object detection methods based on deep learning have the problem of lack of data. In order to solve the data shortage problem, they collect foreign object data of power lines from all different places, use it to train the network, and apply it to other target scenarios that they want to monitor , there are two problems in this approach: (1) Migration problem; the foreign object detection network we trained on various source domain data, put it in other target scenes, the detection performance will be greatly reduced; (2) even if we collect all The data of the non-target domain is used for training, and they still cannot get enough training data

Images