Untreated wounds may lead to serious complications, underscoring the need for consistent monitoring. In clinical practice, wound imaging offers a potential tool for objective assessment. However, obtaining quantitative analysis requires accurate semantic image segmentation, a task that is often time-consuming. Despite deep learning has shown effectiveness in image segmentation, several challenges remain, especially in occluded or low-light images. This study aims to provide a novel semantic segmentation approach for wound images, designed to overcome some of these limitations. A semantic segmentation framework that combines a Convolutional Neural Network-based encoder with a lightweight multilayer perceptron decoder was proposed. A dataset consisting of 451 images was employed for the training, validation, and testing of a semantic segmentation architecture. During the training phase, the model achieved a pixel accuracy of 99.02%, with 98.82% on the validation set and 98.27% on the test set. The corresponding F1 scores were 0.97 for both training and validation, and 0.95 for the test set. The deigned framework and the produced model provided a valuable solution for automatic wound assessment through precise image segmentation approach. The proposed method offers a remarkable contribution to telemedicine, providing a significant advancement for automatic wound management.
