Target detection method based on improved YOLOv8-DES model

CN119478351APending Publication Date: 2025-02-18SHANGHAI NORMAL UNIVERSITY
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411511793.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-28
Publication Date
2025-02-18

AI Technical Summary

Technical Problem

Existing deep learning-based object detection methods reduce detection accuracy in complex backgrounds, target-environmental similarity, occlusion, lighting changes, and target diversity and irregularity, and perform poorly in few sample categories.

Method used

Improve the YOLOv8-DES model by introducing a dual convolution module and an efficient multi-scale attention mechanism, combining Slide loss to dynamically adjust the classification loss weight, optimize the backbone network and feature fusion layer, and improve the processing efficiency of the model in complex scenarios.

Benefits of technology

It improves the detection accuracy of the model in complex scenarios, reduces the number of sample categories during model training, and significantly improves the overall performance of the model and its ability to deal with complex detection environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119478351A_ABST
    Figure CN119478351A_ABST
Patent Text Reader

Abstract

The invention relates to a target detection method based on an improved YOLOv8-DES model, and belongs to the technical field of target detection, and the method comprises the following steps: 1, dividing an image into a training set and a test set; 2, fusing a double-convolution module and a high-efficiency multi-scale attention mechanism module to form an improved model; 3, carrying out forward propagation to extract image features; 4, calculating a loss value of the training set image by adopting a loss function; 5, sequentially carrying out forward back propagation from the last layer of the model, modifying a network weight, and minimizing a loss function; step 6, judging whether model training is completed, if so, entering step 7, and if not, returning to step 3; 7, performing forward propagation to extract image features; 8, performing post-processing on the image features; and step 9, outputting a test picture with the detection frame and the confidence coefficient. According to the invention, the processing efficiency of target detection in a complex model scene is improved, and the number of sample categories during model training is reduced.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Cited By

  • Train control on-board equipment identification method and system, storage medium and electronic equipment

    CN120088586A