Iron ore pixel-level grade prediction method and device, storage medium and electronic equipment
By combining a machine learning model with a convolutional autoencoder and a multilayer perceptron, low-dimensional features are extracted from hyperspectral data, which solves the problems of lag and insufficient accuracy in grade zoning of open-pit iron ore mines, and realizes accurate prediction of iron ore pixel-level grade and refined resource development.
Patent Information
- Application Number
- CN202610219515.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-24
- Publication Date
- 2026-07-03
AI Technical Summary
Existing technologies for grade zoning in open-pit iron ore mines suffer from cumbersome and time-consuming sampling and analysis processes, resulting in delayed grade zoning results that cannot support real-time adjustments to mining plans. Furthermore, manual sampling has limited coverage and cannot fully reflect the spatial heterogeneity of ore grades, leading to insufficient zoning accuracy, increased resource waste, and higher mining costs.
A machine learning model combining convolutional autoencoders and multilayer perceptrons is adopted to extract low-dimensional features from hyperspectral data. A weakly supervised training dataset is constructed using an inverse probability weighting strategy and Monte Carlo random sampling. Unsupervised and supervised training are then conducted to achieve pixel-level grade prediction of iron ore.
It simplifies the grade acquisition process, shortens the analysis time, enables real-time support for dynamic adjustment of mining plans, improves the accuracy of grade zoning, reduces resource waste and mining costs, and can comprehensively characterize the spatial heterogeneity of ore grades.