Local field potentials (LFPs) of different brain regions are contingent on their individual state and functions. While features of LFPs have been used to distinguish brain regions, no study has attempted to determine the precision with which LFPs can infer the recording positions across the whole brain. To tackle this problem, we hypothesize that tiny differences in LFPs from brain regions and their subregions are separable by using state-of-the-art deep learning methods. We implemented four models to best learn the invariant representations and overcome the predictable noise sources from electrode implantation, subject variance, and the environment: a linear network model, an AnyNet (a convolutional neural network, CNN), a vision transformer (ViT), and a recurrent neural network (RNN). We used the International Brain Laboratory (IBL) dataset where the neural activities of mice performing a visual discrimination task were recorded by Neuropixels electrodes. We focused on LFPs recorded during quiet wakefulness and found that the overall accuracy of classifying 472 brain regions from all four models are significantly better than chance level (0.2\%), with the AnyNet and ViT achieving comparable performance at around 40\%. We have tested the robustness of this classification in held-out animals and another behavioral state. We have also used a community identification method to cluster brain regions which share similar LFP features. We further hypothesize that training to localize electrode in mice improves electrode localization performance in humans where data collection is restrictive and expensive, a transfer learning approach. Our work shows that resting-state LFP features contain rich information for deep neural networks to classify 472 brain regions across the whole brain. Future applications of this work enable efficient online brain region identification for precise neuronal signal recording, neural prosthetics implantation, and deep brain stimulation in rodents and humans.