Terrain recognition is employed to discern diverse terrains from visual representations. Distinct terrains manifest distinct characteristics, contingent upon the diverse attributes they encompass, such as varied green foliage, valleys, and land elevations. Within this endeavor, our aim is to cultivate a model capable of discerning these attributes and making precise projections regarding the terrain in question. Our study utilizes the “Jackal UGV Dataset” sourced from IEEE Dataport. This dataset comprises RGB images of seven distinct terrain classes, that are asphalt, brick road, grass, gravel, pavement, sand, and coated floors. This enables a thorough examination of terrain classification in computer vision applications. The proposed approach is about the design of a Terrainfier, which is capable of identifying unacquainted terrains based on transfer learning with models like DenseNet-121, MobileNet and ResNet-50. The performance analysis is furnished through popular metrics like confusion matrix and classification accuracy. The interpretability of the trained transfer learning models is also demonstrated through GradientTape generated heatmaps. The proposed approach yields an accuracy of 97.7% with the DenseNet-121 model, 89.1% with MobileNet model, and 35.16% with ResNet-50 model, respectively.