HERITAGE IDENTIFICATION OF MONUMENTS USING DEEP LEARNING TECHNIQUES

Computer Science

A monument is a physical structure built with the purpose of commemorating a specific individual, event, or achievement. These structures hold significance within particular communities due to their aesthetic, historical, political, technological, or architectural importance, becoming integral parts of their culture and heritage. It is crucial to preserve and document these assets and monuments, as they are closely tied to the regions they belong to. To gather information about a specific subject, one common approach is to use the subject as a search query. The field of image recognition has seen significant advancements, thanks to developments in machine learning and deep learning, particularly in computer vision. With the increasing attention given to international landmarks and monuments, there is a growing need to establish a connection between the physical structures and their digital representations, making automated monument recognition essential. Recognizing monuments from images presents challenges because photographs can vary greatly based on the angle and viewpoint from which they were taken. The two-dimensional representation of monuments adds complexity to the task, as structural similarities among monuments can make identification more challenging. While various classifiers have been used for landmark recognition, the specific task of monument identification has received limited attention, leaving room for improvement. This study's primary goal is to classify images of monuments accurately, utilizing various deep learning architectures to achieve the highest possible accuracy. Several deep learning models, including Canny, Median Filter, and R-CNN, were found to be effective in recognizing monuments. Additionally, preprocessing plays a vital role in converting the dataset into a usable format and improving system performance. Monuments are three-dimensional structures, and differences in image perception can arise due to their varying perspectives. To address this, data augmentation techniques, such as random flip, random rotation, random translation, random zoom, random contrast, random hue, random brightness, and random saturation, were employed to reduce perspective variability. The study evaluated the performance of different design strategies to identify the approach that achieves the highest accuracy in monument recognition.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.