This paper presents a method for automated segmentation of wound areas and scale markers in medical images using a U-Net-based neural network. The model was trained on a dataset of 371 original images, augmented through rotations to increase its generalization ability. Training was conducted over 100 epochs, with dynamic learning rate adjustments to stabilize the process. The model’s performance was evaluated using accuracy, loss metrics, and confusion matrices, with segmentation quality assessed through Dice-Sørensen and Jaccard indices. Results show good performance in segmenting wound areas and scale markers, with the model achieving a median Dice-Sørensen coefficient of 0.7068 for wounds and 0.9397 for scale markers. Future work may include expanding the training dataset and exploring the use of classification-oriented neural networks as encoders to further improve segmentation accuracy.