Diabetic retinopathy (DR) is a leading cause of blindness and visual impairment, particularly among working-age adults, with traditional fundus cameras offering limited retinal coverage. Ultra-widefield (UWF) fundus imaging addresses this by capturing a 200° retinal field, but its larger and more complex images present challenges for automated analysis. This study developed a ResNet-34-based approach for DR classification, employing advanced preprocessing techniques such as contrast enhancement, gamma correction, and cropping, along with data augmentation to enhance generalization. Evaluated on the MICCAI UWF4DR 2024 dataset, the model achieved an average AUC-ROC of 0.7898 through ten-fold cross-validation and 0.8582 on the test set in the challenge using a pretrained ResNet-34. These findings underscore the potential of preprocessing and augmentation strategies in advancing automated DR detection, with future work focused on exploring alternative architectures and optimizing clinical applicability.