A detection method suitable for high-resolution power inspection images and multi-scale devices

By designing a high-resolution image multi-scale target processing framework and detection network, and combining a bounding box protection cutting algorithm with multi-branch convolution and attention modules, the adaptability and inference efficiency and accuracy problems of multi-scale target detection in high-resolution power inspection images are solved. A balance between low inference latency and high detection accuracy is achieved, which is suitable for power equipment inspection by UAV.

CN122176272APending Publication Date: 2026-06-09ZHEJIANG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
ZHEJIANG UNIV
Filing Date
2026-02-06
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Existing detection methods are difficult to adapt to multi-scale targets in high-resolution power line inspection images, and it is difficult to balance inference efficiency and detection accuracy, which cannot meet the real-time requirements of UAV onboard terminals and the efficient processing tasks of servers.

Method used

We designed a high-resolution image multi-scale target processing framework, which combines a bounding box protection cutting algorithm with a detection network consisting of multi-branch convolution and attention modules. Through multi-scale dataset training and an optimal scale filtering mechanism, we achieved accurate detection of multi-category, multi-scale power equipment and defects in high-resolution inspection images.

Benefits of technology

It achieves accurate detection of targets at multiple scales within the range of 40 to 2000 pixels, reduces inference costs, improves detection accuracy, shortens inference latency, adapts to the target distribution characteristics of high-resolution inspection images, and meets the intelligent needs of actual inspection scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122176272A_ABST
    Figure CN122176272A_ABST
Patent Text Reader

Abstract

This invention discloses a detection method adapted to high-resolution power line inspection images and multi-scale equipment, belonging to the fields of machine vision and power line inspection. The method includes: First, combining a bounding box protection cutting algorithm with area-based NMS deduplication, multi-scale sliding window cropping is performed on the acquired power transmission and distribution images to construct a multi-scale dataset; then, based on the YOLO11 structure, multi-branch convolution and multi-branch attention are introduced to construct a detection network, which is trained using the multi-scale dataset; finally, for the image to be detected, multi-scale images are generated according to a scaling factor and input into the trained detection network to complete inference, further employing optimal scale filtering and fusing the multi-scale inference results. This invention combines a high-resolution image multi-scale processing framework with an efficient detection network structure to achieve accurate detection of multi-category, multi-scale power equipment and defects in high-resolution power line inspection images.
Need to check novelty before this filing date? Find Prior Art