The use of artificial intelligence technology for ore prospecting and prediction has become a frontier area in mineral exploration,solving the problem of insufficient accuracy in geological maps for prospecting and prediction and model training.In this study,the ONE-HOT encoding method is proposed and utilized to grid the geological maps.In association with geochemical data,multi-channel two-dimensional grid window data are used as input,and data augmentation methods such as translation,rotation,and scaling are employed to expand the training.An improved ResNet neural network prediction model is used and optimizes hyperparameters through numerous experiments and comparisons.Ore prospecting and prediction was conducted in the Shiquan area,and the results show that the prospective area identified by this method covers 6.5 % of the entire region,including all known gold deposits.Additionally,8 new prospective areas without known gold deposits were predicted.The proposed ONE-HOT encoding method effectively utilizes existing geological map data,and the ResNet neural network prediction model performs significantly better than typical CNN network models and weight of evidence methods.