Using transfer learning to determine the type of mathematical fractals image of Islamic geometric patterns

Abstract Islamic geometric patterns (IGPs) are one of the most elegant forms of art that reflect the Islamic cultural and religious heritage. These patterns are characterized by their complexity and precision. Which makes them a source of inspiration for the design of decorations in light of technological progress. Geometric patterns have never been distinguished before using deep Learning. This data is the first to be used and classified. The research includes classifying IGP into 8 classes Arabesque, Tessellation, Euclidean tiling by convex regular polygons, Koch snowflake, logarithmic spiral, Mandelbrot set, Pythagorean Fractal Tree, Sierpinski Triangles using Transfer Learning. Six deep Learning modes are used VGG19, Mobilenetv2, inception_resnet_v2, xception, NASNetLarge, and Rasnet v2. Adjusting the initialized performance parameters for all model to be able to compare between them. The evaluated performance parameters are training accuracy and loss, evaluation accuracy and loss, precision, recall, confusion matrix, and F1 score. Rasnetv2 gives the higher accuracy, F1 score at small size data.