Version_1.py is a methods of implementing border detection and colour detection. It used Tensorflow as framework, over 300 artifact and real wound images as training materials to create a model that is able to identify wound's border and its change.

#Version_1.py Version_1 used TensorFlow with Kera, U-net architecture with VGG16 encoder

To run the code: Install required libraries first: ''' pip install numpy opencv-python tensorflow requests pillow scikit-learn scikit-image matplotlib ''' Then: ''' python version_1.py '''

'version_1.py': Main program 'v1_border': functions related to border identification 'v1_colour': functions related to colour handling 'v1_evaluation': functions realted to evaluate training performance

'json_images' folder: images with masking in JSON format 'json_masking' folder: JSON format masking 'jj_images' folder: images with masking in .jpg format 'jj_masking' folder: .jpg format masking 'evaluation' folder: images for evaluating model's performance

'fake_jj' folder: JSON format masking 'fake_wound' folder: fake images with .jpg format maksing 'fake_evaluation' folder: fake evaluation images

'wound_segmentation_model.h5': the model file

pip install opencv-contrib-python

All training materials need to add mask manually using Labelbox

The mask that is ready for use

Wound image, model's identification result, and the result after fine tune

Evaluation dataset samples

different type of training materials used

Training and validation loss curve

In case of complicated wound, doctor have the option to circle out wound's border manually, blue line is the border marked manually by a doctor, and the black and white images show the identification result by our algorithms.