As shown in Figure 3, this research utilizes C# programming language to develop a human-computer control interface to facilitate users to operate and set parameters of the imaging system. The interface can let users choose communication ports, adjust LED brightness, set access paths, and imaging modes. In addition, all data collected by the imaging system is uploaded to the cloud database through the Internet, and the data is automatically classified and stored by using MySQL for subsequent analysis. Also, PHP is used to link databases to display wound information on the web platform, which can let patients and medical personnel access data by using mobile devices.

After obtaining the wound image, the first step is to detect the location of the wound, which facilitates subsequent analysis. We utilize the U-Net model for wound area segmentation, as depicted in the top-left image of Figure 4. The dataset comprises 810 images from open-source databases and 136 images from National Cheng Kung University Hospital. The data is split into a training set and a testing set, with a ratio of 8:2. The image on the right side of Figure 4 shows the model's segmentation results, where the green mask represents the segmented wound area. The model achieves a Precision of 91.9%, Recall of 91.2%, and Dice coefficient of 91.5%.

After segmenting the wound area, the next step involves wound tissue classification. Wound tissues are categorized into granulation tissue, slough tissue, and eschar tissue. Granulation tissue indicates active wound healing, resembling granules of flesh growing within the wound. A higher presence of granulation tissue signifies better healing capability. Eschar tissue indicates tissue that has lost its ability to regenerate, while slough tissue is affected by excessive bacterial growth, hindering wound closure. Both necrotic and slough tissues require debridement for removal. Therefore, identifying wound tissues quantifies the wound's healing potential, aiding decisions on debridement necessity or changes in care methods.

The classification results are shown in Figure 5. The red areas represent granulation tissue, the yellow areas denote slough tissue, and the black areas indicate eschar tissue. The wound tissue classification model achieves a Precision of 85.5%, Recall of 85%, and Accuracy of 85%.

Figure 6 shows photos of a patient's wound at different time points before and after debridement that were performed at National Cheng Kung University Hospital. Observations show that oxygen levels in the wound area are higher than those in the surrounding skin. Before debridement, images reveal extensive necrotic areas with lower oxygen levels compared to other parts of the wound. After debridement, there is a noticeable increase in oxygen levels in these areas, indicating improved healing capability post-debridement. Our system captures the oxygen distribution within wounds, providing physicians with critical wound information to assist in debridement and care decisions.