Remote sensing image (RS) classification has a broad range of potential applications in various fields. However, remotely sensed images comprise various land cover elements arranged in intricate spatial patterns, presenting a challenge for accurate recognition. Fortunately, breakthroughs in deep learning techniques brought new opportunities for revolutionizing RS image processing. The convolutional neural network (CNN), as a deep learning technique, has garnered considerable attention in recent years, particularly for its widespread application in remote-sensing image scene classification. Recent advancements have demonstrated its effectiveness in this domain. This study uses CNN as the basis for the land-use scene classification task. We chose the UC-Merced dataset as the experimental object, which contains 21 different categories of land-use scene images. Then, we leverage a modified version of the ResNet50 model to overcome the feature degradation problem in the deep learning process. The experimental results show that ResNet50 with transfer learning exhibits the best performance in the land-use scene classification task, and its accuracy (95.29% (±0.66%)), precision (95.35% (±0.68%)), F1-score (95.23% (±0.67%)), and Kappa coefficient (95.05% (±0.70%)) are all better than the other models. In conclusion, this study serves as a valuable reference for selecting deep learning models for land-use scene classification tasks and offers insightful guidance for future research and applications in related domains.