一种基于边缘计算和注意力感知压缩的图像检测方法
By building an attention-aware compression image detection system in an edge computing environment, and optimizing the compression ratio and model selection using the attention-aware module and policy generation module, the resource and bandwidth constraints of image detection tasks in edge computing are solved, achieving low-latency and high-precision image detection results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CENT SOUTH UNIV
- Filing Date
- 2026-01-23
- Publication Date
- 2026-07-17
AI Technical Summary
In edge computing environments, image detection tasks face challenges such as limited computing resources and insufficient network bandwidth. Existing methods fail to effectively combine the importance of different regions of an image with the selection of inference models, resulting in loss of inference accuracy and low resource utilization.
An image detection system based on edge computing and attention-aware compression is constructed. The key regions of the image are extracted through the attention-aware module. The compression rate and inference model selection are optimized by combining the image segmentation module and the policy generation module. An edge-end collaborative architecture is adopted, and attention weight maps are generated using the MobileNet V3 classification network and GradCAM++ technology for adaptive compression and model optimization.
It achieves low-latency, high-precision image detection under resource-constrained conditions. Through attention-driven adaptive compression and joint optimization strategies, it improves system utility and adapts to dynamically changing edge computing environments.
Smart Images
Figure CN121937846B_ABST