Water surface object recognition system and method and application based on 5G MEC

Abstract

The invention discloses a water surface object recognition system and method and application based on 5G MEC. The method comprises the steps of completing a water surface object automatic detection and recognition model based on an improved Yolov4 deep learning method through an edge cloud in advance, temporarily storing model parameters, and making a new image sample through the returned data, so as to update the model, and achieve transfer learning; and finally, sending an identification result to an unmanned ship control system and a user side. The system comprises an unmanned ship and an edge cloud module. In the application of water surface object recognition based on 5G MEC, video data are converted into video code stream data, the video code stream data are uploaded to edge cloud by adopting a 5G network communication module, and the edge cloud is used for detecting and recognizing a water surface moving target. According to the scheme, YOLOv4 is combined to serve as a target detection algorithm, the optimal balance between the detection precision and efficiency is achieved, the depth and transfer learning recognition algorithm is combined, the obstacle avoidance capacity of the unmanned ship is enhanced, and the garbage collection efficiency is improved.

Description

technical field [0001] The present invention relates to the technical field of intelligent recognition, in particular to a system and method for water surface object recognition and an application based on 5G MEC. Background technique [0002] At present, there are many kinds of garbage on the surface of the water. In order to improve the functionality and safety of the clean unmanned ship, the unmanned ship needs to judge whether it is garbage collection or obstacle avoidance when sailing, and the existing unmanned ship lacks a surface object recognition system; existing The ground station of the clean unmanned ship lacks a high-definition video return system and a visualization platform, and cannot provide real-time and intuitive feedback of surrounding environment information to users, resulting in poor user experience in deployment configuration. Therefore, how to quickly realize water surface object recognition and intuitive feedback It is very necessary for users; [...

AI Technical Summary

Problems solved by technology

[0002] At present, there are many kinds of garbage on the surface of the water. In order to improve the functionality and safety of the clean unmanned ship, the unmanned ship needs to judge whether it is garbage collection or obstacle avoidance when sailing, and the existing unmanned ship lacks a surface object recognition system; the existing The ground station of the clean unmanned ship lacks a high-definition video return system and a visualization platform, and cannot provide real-time and intuitive feedback of surrounding environment information to users, resulting in poor user experience in deployment configuration. Therefore, how to quickly realize water surface object recognition and intuitive feedback It is very necessary for users;

[0003] At the same time, when recognizing a video, each frame is input as a separate image, and independent recognition operations are performed separately. This strategy makes it impossible to use the previously calculated results for each frame of recognition. , a lot of repetitive work has been done in the process of recognizing video; in addition, if the deep learning method is used for the recognition of water objects, although the detection accuracy has achieved good results, it requires a large amount of data and a variety of training parameters, and for unmanned For boats, it is difficult to obtain complex failure case data and conduct training process

In addition, during the driving process of the unmanned boat, the detection and recognition of the target is based on the dynamic background. The traditional image recognition algorithm has a large number of image preprocessing and feature extraction links, which will increase the calculation time.

Images