Image instance detection and segmentation model construction method based on edge information enhancement

By enhancing the image instance detection and segmentation model with edge information, combined with ResNet and bidirectional feature pyramid network, the problems of unclear boundaries and inaccurate segmentation of scale-diverse targets in remote sensing and aerial images are solved, achieving higher-precision instance detection and segmentation.

CN118762042BActive Publication Date: 2025-09-23HUAZHONG UNIV OF SCI & TECH +1
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202410736505.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-06-07
Publication Date
2025-09-23
Estimated Expiration
2044-06-07

AI Technical Summary

Technical Problem

Existing technologies have the problem of low accuracy in remote sensing and aerial images, especially in the segmentation of image instances with unclear boundaries and scale-diverse targets. In particular, the semantic features of small targets are easily interfered with or obscured by the background, resulting in inaccurate segmentation.

Method used

An image instance detection and segmentation model based on edge information enhancement is adopted. Through the feature extraction module, edge information enhancement module and multi-scale feature fusion module, ResNet and bidirectional feature pyramid network are combined to extract and enhance edge features, and perform multi-scale adaptive fusion and prediction.

Benefits of technology

It improves the accuracy of image instance detection and segmentation, can more accurately locate and segment target edges, enhances edge perception capabilities, adapts to different task requirements, and improves the robustness of the model and the refined detection and segmentation effects.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118762042B_ABST
    Figure CN118762042B_ABST
Patent Text Reader

Abstract

The present invention belongs to the technical field of image instance detection and segmentation, and discloses a method for constructing an image instance detection and segmentation model based on edge information enhancement. The method comprises: using a dataset to train an instance detection and segmentation network to obtain an instance detection and segmentation model; the instance detection and segmentation network comprises: a feature extraction module for extracting high- and low-resolution feature maps of different scales from each input sample image to form a feature pyramid; an edge information enhancement module for performing edge information enhancement on the input sample image to obtain an edge information-enhanced feature map; a multi-scale feature fusion module for performing multi-scale adaptive fusion of the edge information-enhanced feature map and other feature maps in the feature pyramid to obtain a fused feature map; and a prediction module for predicting the instance category of the target, the bounding box position of the instance, and the binary mask based on the fused feature map. The present invention can improve the accuracy of image instance detection and segmentation.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Instance segmentation method fusing hole convolution and edge information

    CN110348445A

  • Remote sensing image lightweight semantic segmentation method based on edge decoupling

    CN113159051A