Multi-spectral imaging significantly expands the capabilities of conventional imaging technology by extracting rich spectral information from images. Multi-spectral imaging is frequently employed in agriculture, the military, health, industry, and meteorology. Multi-spectral images have redundant information;thus, preprocessing is required to decrease the dimension. Most researchers now use preprocessing techniques before classifying data in recent years. The merits and disadvantages of standard dimensionality reduction techniques are first described and addressed. These techniques are founded on principles.Then, the features and application areas of classical and deep learning methods for categorization are examined. In contrast, the latter has good adaptability and high classification accuracy, while the former is more affordable and has more developed theories. Currently, approaches could be improved in order to use spectral data effectively and conserve computational resources. Traditional techniques will be enhanced and employed extensively in the future, while noveldesigns with greater flexibility and precision will be created