Classification of Defective Intensity Levels of Paint in Heritage Buildings using the CNN-SVM Technique
Pritha Singha Roy, Vinay Kukreja, Vishal Jain, Satvik Vats
Abstract
For three-dimensional (3D) cultural heritage objects paintings that are destroyed or dented and can no longer be 3D scanned to establish digital archives, this study suggests a deep learning-based hybrid method. This study discusses about cultural heritage paintings' intensity levels and healthy painting levels. To differentiate, Convolution Neural Networks are employed to extract the feature of paintings. If there are visible application flaws or imperfections such as acrylic runs, shrinks, freckles, silt, empty or deprived adorned spaces, color variations, delamination, abrasive and random squiggles, abrasive scars, pustules, consistency of texture tier, or other grooves, the exterior is considered to be defective. With the help of CNN and SVM, the intensity levels are recognized. Initially, the image classification algorithm's recognition accuracy is 90.79%. Second, an object detection algorithm may achieve a recognition accuracy of 95.61 percent, proving its accuracy and effectiveness. The entire piece of code was written in Python and executed there, along with libraries for TensorFlow and Keras. The deep CNN model's highest accuracy, which has been successfully assembled is 84.23%. The average precision value of Healthy painting is 83.95, Intensity level 1 is 84.01, Intensity level 2 is 85.14, Intensity level 3 is 83.14, Intensity level 4 is 84.94, and Intensity level 5 is 84.22.