Deep Learning Applications in Automatic Interpretation of Nigerian Aeromagnetic Data: Training, Validation, and Deployment for Geological Mapping

The Nigerian Geological Survey Agency holds a national aeromagnetic dataset covering the entire country at varying flight-line spacings, representing a vast geophysical resource for geological mapping that is currently under-interpreted due to the time and expert-knowledge demands of conventional manual interpretation. This research develops, trains, and validates a deep learning system for the automatic geological interpretation of Nigerian aeromagnetic data, addressing the specific challenges of Nigerian basement complexity, multi-age sedimentary cover variability, and the heterogeneous quality of different vintages of the national dataset. A convolutional neural network architecture is developed for multi-class geological unit discrimination from aeromagnetic enhancement images, trained on a supervised dataset of 15,000 image patches labelled with geological unit types derived from the 1:250,000 Nigerian geological map series. A second recurrent neural network branch processes structural lineament time-series derived from multidirectional derivative filters to improve structural boundary delineation. The combined network achieves a geological unit classification accuracy of 83 percent on a held-out test set drawn from four geologically diverse states. Application to previously unmapped areas in Kebbi and Sokoto states produces geological maps whose unit boundaries are confirmed at 76 percent of accessible verification outcrop localities. A transfer learning experiment demonstrates that retraining the network with 500 new labelled samples from a different geological province improves accuracy in that province from 61 to 79 percent. The research provides a deployable deep learning system for accelerating the geological mapping programme of the Nigerian Geological Survey Agency. Keywords: deep learning, aeromagnetic interpretation, geological mapping, convolutional neural network, Nigeria