Practical application of semantic segmentation to extract elements from landscape images

An example of object detection using YOLOv8

Abstract

In this report, among the various approaches to object detection, I focus on semantic segmentation, which enables object detection on a pixel-by-pixel basis.Specifically, by using YOLOv8 as an example, I clarify whether image analysis techniques that use low-code programming can be fully understood and utilized even by non-experts. The purpose of this study was to clarify the path to the social implementation of object detection in landscape urban development, as well as to examine various points to note in image analysis and identify issues that may arise in future implementation.I was able to fine-tune three pretrained YOLOv8 models and verify both the accuracy of their additional training and their ease of use in the implementation procedure for object detection. Because of space constraints, this report omits the fine-tuning of detailed object detection, but I found that approximately 9 h of additional learning helped significantly to improve the accuracy of detection of landscape elements.However, it also became clear that there are still many hurdles to overcome before object detection can be easily implemented and established as a mature analytical method. In future investigations, I plan to continue reviewing further related research on object detection and to seek ways to refine and simplify image analysis methods that will contribute to landscape urban planning.