Remote sensing image building identification method and system based on multi-neural network integration, and electronic equipment
By constructing a multi-neural network integrated remote sensing image building recognition method, the problems of high computing power, large parameters, slow recognition and lack of rules in edge devices are solved, and efficient and accurate building recognition and safe flight strategies are achieved.
Patent Information
- Application Number
- CN202510735486.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-04
- Publication Date
- 2025-09-23
AI Technical Summary
Existing remote sensing image building recognition models require high computing power and are not suitable for edge device deployment. They have large model parameters, slow recognition speed, poor regional information perception capabilities, and a lack of rule-based strategies, resulting in high potential risks for edge devices.
A primitive edge network consisting of a Mobilenet encoder, a Transformer encoder, a U-Net decoder, and the output of a segmentation head is constructed and trained on conventional and single-class building datasets. The model parameter package is stored on a cloud server, and the model is optimized based on the edge device positioning and regional information, and recognition is performed by matching spatial domain rules.
It reduces model parameters and computing power requirements, improves building recognition accuracy, implements adaptive matching and hard limits, and reduces potential risks of edge devices.