Terrain recognition is a vital addition that would advance the functionality of autonomous systems such as self-driving cars, mobile robots, and environmental monitoring platforms. Ineffective terrain classification, systems will change navigation approaches for any terrain to avoid dangerous terrain types and save resources by having more efficient journeys. The objective of this paper was to consider deep learning-based methods in recognizing four basic categories of terrains: sandy, marshy, rocky, and grassy. These architecture-based CNNs, namely VGG-16, EfficientNet, and InceptionResNetV2 were used for terrain classification. The results shown the comparison between the models indicating which deep learning models is powerful for different terrain classifications, opening opportunities for improving the performance of autonomous systems in complex and different environments.