Accurate wound image analysis is important for effective treatment planning and monitoring healing progression. This paper introduces a deep learning methodology for automated wound segmentation by developing an innovative framework that uses EfficientNet and UNet architectures. Our research tackles essential issues in wound assessment, which involve multi-class segmentation along with boundary precision and test-time augmentation. To enhance segmentation quality, the proposed methodology employs a composite loss function combining focal, boundary-aware, and dice loss modules. The experimental results show equivalent or better performance than current methodologies with our three-class model achieving an average F1 score of 0.8876. This work represents a significant advancement toward comprehensive wound assessment systems that can support decision-making across diverse healthcare settings.