Target intelligent matching and identification method based on multi-dimensional image

Abstract

The invention discloses a target intelligent matching and identification method based on a multi-dimensional image. The method specifically comprises the following steps: S1, screening and arranging images collected by a fixed camera, a robot, and an unmanned plane camera in a power grid deployment and control inspection region, and forming corresponding multi-scale target training data; S2, proposing a matching degree maximization criterion of the same target under multiple scales; S3, providing a multi-dimensional image joint learning model; S4, realizing target matching with different resolutions under different cameras; and S5, further identifying the successfully matched images. Expression information of the target under the low resolution is effectively enhanced through information of different scales, meanwhile, the same target under the different resolutions is judged through the matching degree maximization criterion, then different-scale optimal measurement models are learned, and compared with a single-scale matching scheme, the method has the advantage of keeping higher matching accuracy in various complex environments, and the successfully matched target can be further identified.

Description

technical field [0001] The invention relates to computer vision technology, in particular to an intelligent target matching and recognition method based on multi-dimensional images, which combines multi-dimensional images to achieve the purpose of improving target matching accuracy and recognition accuracy, thereby improving power grid inspection efficiency. Background technique [0002] With the continuous growth of the national grid line mileage, the risk factor is high, and the time-consuming and labor-intensive traditional manual inspection can no longer meet the development needs. Machines have been widely used in power grid inspection. This intelligent inspection method not only has low risk factor and low cost, but also is less affected by environmental and weather factors. important direction. [0003] However, while intelligent inspection has broad prospects, it is limited by the development of technology, and there are also some defects that cannot be ignored. For...

AI Technical Summary

Problems solved by technology

However, deep learning requires large-scale data to train a model with high accuracy. In addition, the environment of the inspection range is complex, the images taken by the cameras carried by robots and drones are not clear enough, and the target features are not obvious, which seriously affects the recognition accuracy.

Images