A rope detection and removal method based on deep learning

Abstract

The invention discloses a method for rope detection and rope removal based on instance segmentation, comprising the following steps: S1, collecting image data photographed by an airship pod and manually photographed wire image data, and preprocessing the image data; S2, constructing Configurable deep learning model, initialize the model parameters, divide the image data into training set and test set, and use the data in the training set and test set to train the deep learning model; S3, the image to be processed, use the trained model. Process to obtain the preliminary segmentation map of the rope; S4, convert the preliminary segmentation map into a grayscale image, and use the maximum inter-class variance algorithm to finely segment the rope; S5, use the fast multi-pole algorithm to remove the rope from the image after the finely segmented rope and fix the image. Combined with the semantic segmentation algorithm, the invention provides a method for segmenting and removing ropes in aerial images under complex backgrounds.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a rope detection and removal method based on deep learning. Background technique [0002] Guangqi Yunyun is a special helium-filled fiber capsule. Its main working range is in the airspace of 4,000 to more than 5,000 meters high. Monitoring and aerial photography services are performed through the pod below the airship. The pod is flexibly equipped with a variety of functional modules. Optoelectronic composite cables are connected to the big data center. In actual aerial photography, the obtained image will contain rope information, and the rope with this non-uniform structure is not only not conducive to scene analysis, but also to the detection of effective targets. Therefore, image information processing technology must be used to eliminate rope interference and restore The actual scene in the image provides data support for further analysis of the image. [0003] T...

AI Technical Summary

Problems solved by technology

Although the accuracy and real-time performance of detection and segmentation are improving step by step, the deep learning algorithm will have great redundancy and limitations in instance segmentation, and there is a lot of noise interference in the video images monitored on the Kuangqi cloud account. The existing model has the problem of not being able to converge in the case of small target segmentation and small amount of data.

Images