This chapter presents a comprehensive introduction to the foundational concepts of deep learning applied to hyperspectral image analysis. It begins by outlining the core components of neural networks, including artificial neuron models, backpropagation algorithms, and essential architectural elements such as fully connected layers, convolutional filters, and pooling operations. Building upon these fundamentals, the chapter explores the evolution of deep neural architectures tailored to hyperspectral imaging analysis, focusing on Convolutional Neural Networks for spatial-spectral feature extraction, Recurrent Neural Networks for modeling spectral sequences, and attention-based models such as transformers for capturing long-range dependencies. Recent advances are also discussed, including state-space models such as Mamba and interpretable paradigms such as Kolmogorov-Arnold Networks. Special emphasis is placed on the unique challenges associated with hyperspectral data, including high spectral dimensionality, spatial heterogeneity, scarcity of labeled samples, and the need for effective spectral-spatial fusion. The chapter concludes by framing these challenges as opportunities for architectural innovation and highlights promising directions for future research in deep learning for hyperspectral image analysis.