This work assesses the application of a two-stage deep learning pipeline for the self-supervised pretraining and supervised fine tuning of winter crop and weed image classification. The experiments on the WinterCropWeedDB dataset are conducted using a single internal split, with Grad-CAM visualizations included.