딥러닝 기반의 고해상도 이종 원격탐사 상 간 구름탐지 및 복원

조수민,송준영,원태연,어양담 한국측량학회 2022 한국측량학회지 Vol.40 No.6

This paper proposed a method of restoring a cloud occluded area by a deep learning method by detectingclouds from high-resolution satellite images and learning between heterogeneous satellite images. After trainingaerial images with clouds, clouds in the Worldview-3 image were detected with YOLOv3 and the restorationarea was designated. CycleGAN was used among deep learning techniques for the restoration of the detectedcloud occlusion area, and an image restoration experiment was conducted through learning between the opticalheterogeneous satellite images, Worldview-3 and Kompsat-3 images. As a result, our proposed methodologyderived a decrease in RMSE (Root Mean Square Error) by about 72.3% compared to the result of applying theHistogram Matching Method, and a high R² of about 0.98. Through this, it was confirmed that cloud detectionin high-resolution satellite images is possible by learning aerial photographs, and cloud occlusion areas can berestored through learning between heterogeneous satellite images. 본 논문은 고해상도 위성영상에서 구름을 탐지하고, 이종 위성영상 간 학습하여 딥러닝 방식에 의한 구름 폐색영역 복원 방법을 제안하였다. 구름이 있는 항공사진을 학습한 후 YOLOv3로 Worldview-3 영상 내 구름을 탐지하였고 복원영역을 지정하였다. 탐지된 구름 폐색영역 복원은 딥러닝 기법 중 CycleGAN을 사용하였고, 광학 이종 위성영상인 Worldview-3 영상과 Kompsat-3 영상 간의 학습을 통해 영상복원 실험을 수행하였다. 실험 결과, 제안한 방법론은 히스토그램 매칭 기법 적용 결과에 비해 RMSE 값이 약 72.3% 감소하였으며, R²는 약 0.98로 높게 도출되었다. 이를 통해, 항공사진을 이용한 고해상도 위성영상 내 구름 탐지 및 이종 위성영상 간 학습을 통한 구름 폐색영역복원 가능성을 확인할 수 있었다.

YOLOv3 객체 검출을 이용한 AR 관광 서비스 프레임워크

김인선,정치서,정계동 한국인터넷방송통신학회 2021 한국인터넷방송통신학회 논문지 Vol.21 No.1

교통 수단과 모바일의 발전으로 관광 여행 수요가 증가하고 관련 산업 또한 크게 발전하고 있다. 디지털 미디어 기술 중 한 분야인 증강현실과 관광 콘텐츠의 접목 또한 활발하게 연구 중이며 인공지능은 이미 관광 산업과 다양한 방향으로 접목되어 관광객의 여행 경험을 풍부하게 만들어준다. 본 논문에서는 관광지역을 축소해 제작한 미니어처 모형 을 스캔하면, 사전에 딥러닝을 이용해 학습된 모델을 기반으로 해당 관광지를 찾은 뒤 관련 정보와 3D 모델을 AR 서비 스로 제공하는 시스템을 제안한다. 다양한 딥러닝 신경망 중 하나인 YOLOv3 신경망을 사용해 모델 학습과 객체 검출을 진행하므로, 빠른 속도로 물체 검출이 이루어져 실시간으로 서비스를 제공할 수 있다. With the development of transportation and mobiles demand for tourism travel is increasing and related industries are also developing significantly. The combination of augmented reality and tourism contents one of the areas of digital media technology, is also actively being studied, and artificial intelligence is already combined with the tourism industry in various directions, enriching tourists' travel experiences. In this paper, we propose a system that scans miniature models produced by reducing tourist areas, finds the relevant tourist sites based on models learned using deep learning in advance, and provides relevant information and 3D models as AR services. Because model learning and object detection are carried out using YOLOv3 neural networks, one of various deep learning neural networks, object detection can be performed at a fast rate to provide real-time service.

Performance Evaluation of YOLOv3 and YOLOv4 Detectors on Elevator Button Dataset for Mobile Robot

Sumaira Manzoor,Eun-Jin Kim,Gun-Gyo In,Tae-Yong Kuc 제어로봇시스템학회 2021 제어로봇시스템학회 국제학술대회 논문집 Vol.2021 No.10

The performance evaluation of an AI network model is the important part for building an effective solution before its deployment in real-world on the robot. In our study, we have implemented YOLOv3-tiny and YOLOv4-tiny darknet based frameworks for performance evaluation of the elevator button recognition task and tested both variants on image and video datasets. The objective of our study is two-fold: First, to overcome the limitation of elevator buttons dataset by creating new dataset and increasing its quantity without compromising the quality; Second, to provide a comparative analysis through experimental results and the performance evaluation of both detectors using four machine learning metrics. The purpose of our work is to assist the researchers and developers in decision making of suitable detector selection for deployment in the elevator robot towards button recognition application. The results show that YOLOv4-tiny outperforms YOLOv3-tiny with an overall accuracy of 98.60% compared to 97.91% at 0.5 IoU.

Fault Detection in 3D Printers using an Improved YOLOv5 with Hyperparameter Tuning

Made Adi Paramartha Putra,Mark Verana,Revin Naufal Alief,Dong-Seong Kim,Jae-Min Lee 한국통신학회 2021 한국통신학회 학술대회논문집 Vol.2021 No.11

In this study, a modification of you only look once version 5 (YOLOv5) is presented to detect an error during the printing process of the fused filament fabrication (FDM) 3D printer. The improvement has been made in hyperparameter selection. The existing YOLOv5 uses the COCO dataset as the default hyperparameter, which is unsuitable for the 3D printing process. Therefore, we captured the image dataset using an FDM 3D printer to improve the detection result generated by the YOLOv5 model. The results show that the improved YOLOv5 with hyperparameter tuning is better than the traditional version of YOLOv5 based on the mean absolute precision (mAP).

딥러닝 사물 인식 알고리즘(YOLOv3)을 이용한 미세조류 인식 연구

박정수 ( Jungsu Park ),백지원 ( Jiwon Baek ),유광태 ( Kwangtae You ),남승원 ( Seung Won Nam ),김종락 ( Jongrack Kim ) 한국물환경학회(구 한국수질보전학회) 2021 한국물환경학회지 Vol.37 No.4

Algal bloom is an important issue in maintaining the safety of the drinking water supply system. Fast detection and classification of algae images are essential for the management of algal blooms. Conventional visual identification using a microscope is a labor-intensive and time-consuming method that often requires several hours to several days in order to obtain analysis results from field water samples. In recent decades, various deep learning algorithms have been developed and widely used in object detection studies. YOLO is a state-of-the-art deep learning algorithm. In this study the third version of the YOLO algorithm, namely, YOLOv3, was used to develop an algae image detection model. YOLOv3 is one of the most representative one-stage object detection algorithms with faster inference time, which is an important benefit of YOLO. A total of 1,114 algae images for 30 genera collected by microscope were used to develop the YOLOv3 algae image detection model. The algae images were divided into four groups with five, 10, 20, and 30 genera for training and testing the model. The mean average precision (mAP) was 81, 70, 52, and 41 for data sets with five, 10, 20, and 30 genera, respectively. The precision was higher than 0.8 for all four image groups. These results show the practical applicability of the deep learning algorithm, YOLOv3, for algae image detection.

드론 영상의 YOLO 딥러닝 기법 적용을 통한 개인형 이동장치 탐지

김준석,이태현,염준호 한국측량학회 2023 한국측량학회지 Vol.41 No.4

Recently, the utilization rate of PM (Personal Mobility) and its users has been rapidly increasing as a short distance transportation option. As the consumption patterns in modern cities shift towards the sharing economy, various shared mobility platforms have been developed, leading to the emergence of PM in the form of shared electric scooters. Consequently, there has been a simultaneous increase in companies providing shared PM services. However, due to the diverse types of shared PM offered by different service providers and variations in the number of providers across regions, the comprehensive management of PMs has become more challenging. Therefore, this paper aims to evaluate the feasibility of utilizing the YOLOv3 algorithm to detect shared PM objects from drone images and to assess accuracy, thereby verifying the potential for integrated PM management of PMs. PM images within the experimental area were collected using drones, and individual objects were labeled to train a deep learning model for PM detection. Subsequently, an accuracy evaluation was conducted to validate the feasibility of the approach. The experimental results demonstrated 80% recall and 87% precision accuracy, and an AP (average precision) value of 0.73, confirming the viability of utilizing the YOLOv3 algorithm on drone images for PM detection. 최근 단거리 교통수단으로 개인형 이동장치와 이를 사용하는 사용자의 이용률이 빠르게 증가하고 있다. 또한, 현대도시의 소비 형태가 공유경제의 형태로 변화하며 관련 공유 플랫폼이 개발됨에 따라 개인형 이동장치인 PM(Personal Mobility)이 공유 전동킥보드 형태로 나타났으며, 이와 동시에 공유 PM 서비스를 제공하는 업체도 같이 증가하고 있다. 그러나 PM이 서비스 제공 업체마다 종류가 다르고, 지역마다 그 업체의 수가 달라 통합적인 관리가 더욱 어려운 상황이다. 따라서 본 논문에서는 드론을 통해 수집한 영상에서 YOLOv3 알고리즘으로 여러 업체의 PM 객체를 탐지하여, 통합적인 관리의 활용 가능성이 있는지 분석하고 정확도 평가를 수행하였다. 실험지역 내 PM이 포함된 드론 영상을 수집하고 PM 객체를 레이블링하여 딥러닝 모델을 학습시켜 PM을 탐지하였다. 정확도 평가 결과 재현율 80%, 정밀도 87%의 탐지 정확도와 0.73의 AP값을 얻었으며 이를 통해 드론 영상에서 YOLOv3 알고리즘을 활용하여 PM 검출을 수행하는 것이 가능함을 확인하였다.

Object Recognition from Spherical Camera Images Based on YOLOv3

Tomohiro Kai,Humin Lu,Tohru Kamiya 제어로봇시스템학회 2020 제어로봇시스템학회 국제학술대회 논문집 Vol.2020 No.10

The aging of Japan is remarkable, and attention has been focused on the use and utilization of assistive devices. One of them is electric wheelchair, which enables physical disability people to easily operate it using a handle or a joystick. However, accidents are occurring frequently with increasing demand by using electric wheelchair. Therefore, developing an autonomous electric wheelchair is required to reduce accidents such as maneuvering mistakes, reduce the accident rate, improve convenience, and reduce the burden on caregivers. In this paper, we focus on the recognition of obstacles and use panoramic images obtained from a spherical camera that can easily handle information from all directions at low cost. A spherical camera is attached to an electric wheelchair, and images are cut out from the sequential images obtained by running. For image analysis, YOLOv3, which has been successful in the field of image recognition in recent years, is used. In the proposed method, considering the distortion of the image caused by using the spherical camera, the improvement of the model of YOLOv3 is examined, and the validity with the actual data is verified.

백본이 수정된 YOLOv3를 이용한 약용작물 병충해 탐지 AI 서비스 구현

허진경(JinKyoung Heo),이근호(Geunho Lee) 한국정보기술학회 2022 Proceedings of KIIT Conference Vol.2022 No.12

식물이 생산하는 화학 물질은 자신의 생존을 향상하게 하는 과정 과정에서 화학 물질이 더 많이 생산될 수 있다. 특히 약용으로 사용되는 천연 추출물을 다량 함유한 약용작물은 적은 노동력으로 농업 생산성을 최대화할 수 있다. 본 논문은 식물의 천연 추출물 함량을 높일 수 있는 환경적 요인을 알아보기 위한 과정에서 발생할 수 있는 병충해로 인한 피해를 줄이기 위해 인공신경망 객체탐지 알고리즘인 YOLOv3의 백본을 수정한 모델을 적용한 인공지능 모델과 탐지 서비스 구현을 제안한다. Chemicals produced by plants can be produced more in the process of improving their own survival. In particular, medicinal crops containing a large amount of natural extracts used for medicinal purposes can maximize agricultural productivity with less labor. This paper proposes an artificial intelligence model and detection service implementation with a modified model of YOLOv3, an artificial neural network object detection algorithm, to reduce damage caused by pests in the process of identifying environmental factors that can increase the natural extract content of plants.

AI Fire Detection & Notification System

You-min Na(나유민),Dong-hwan Hyun(현동환),Do-hyun Park(박도현),Se-hyun Hwang(황세현),Soo-hong Lee(이수홍) 한국컴퓨터정보학회 2020 韓國컴퓨터情報學會論文誌 Vol.25 No.12

본 논문에서는 최근 가장 신뢰도 높은 인공지능 탐지 알고리즘인 YOLOv3와 EfficientDet을 이용한 화재 탐지 기술과 문자, 웹, 앱, 이메일 등 4종류의 알림을 동시에 전송하는 알림서비스 그리고 화재 탐지와 알림서비스를 연동하는 AWS 시스템을 제안한다. 우리의 정확도 높은 화재 탐지 알고리즘은 두 종류인데, 로컬에서 작동하는 YOLOv3 기반의 화재탐지 모델은 2000개 이상의 화재 데이터를 이용해 데이터 증강을 통해 학습하였고, 클라우드에서 작동하는 EfficientDet은 사전학습모델(Pretrained Model)에서 추가로 학습(Transfer Learning)을 진행하였다. 4종류의 알림서비스는 AWS 서비스와 FCM 서비스를 이용해 구축하였는데, 웹, 앱, 메일의 경우 알림 전송 직후 알림이 수신되며, 기지국을 거치는 문자시스템의 경우 지연시간이 1초 이내로 충분히 빨랐다. 화재 영상의 화재 탐지 실험을 통해 우리의 화재 탐지 기술의 정확성을 입증하였으며, 화재 탐지 시간과 알림서비스 시간을 측정해 화재 발생 후 알림 전송까지의 시간도 확인해보았다. 본 논문의 AI 화재 탐지 및 알림서비스 시스템은 과거의 화재탐지 시스템들보다 더 정확하고 빨라서 화재사고 시 골든타임 확보에 큰 도움을 줄 것이라고 기대된다. In this paper, we propose a fire detection technology using YOLOv3 and EfficientDet, the most reliable artificial intelligence detection algorithm recently, an alert service that simultaneously transmits four kinds of notifications: text, web, app and e-mail, and an AWS system that links fire detection and notification service. There are two types of our highly accurate fire detection algorithms; the fire detection model based on YOLOv3, which operates locally, used more than 2000 fire data and learned through data augmentation, and the EfficientDet, which operates in the cloud, has conducted transfer learning on the pretrained model. Four types of notification services were established using AWS service and FCM service; in the case of the web, app, and mail, notifications were received immediately after notification transmission, and in the case of the text messaging system through the base station, the delay time was fast enough within one second. We proved the accuracy of our fire detection technology through fire detection experiments using the fire video, and we also measured the time of fire detection and notification service to check detecting time and notification time. Our AI fire detection and notification service system in this paper is expected to be more accurate and faster than past fire detection systems, which will greatly help secure golden time in the event of fire accidents.

스테레오 비전과 YOLOv3 를 이용한 드론의 3 차원 실내 위치 추정 알고리즘 개발

변영훈(Younghun Byeon),강민재(Minjae Kang),임현준(Hyeon Jun Lim),김한솔(Han Sol Kim) 제어로봇시스템학회 2021 제어·로봇·시스템학회 논문지 Vol.27 No.11

In this paper, we propose a three-dimensional indoor position estimation algorithm for drones using stereo vision and YOLOv3. First, we find the bounding box of a drone in the image using a deep-learning-based object detection algorithm called YOLOv3. To this end, we collect the training dataset consisting of drone images. In addition, the object detection performance of the YOLOv3 algorithm is improved by dividing object class labels of the same drone based on the angle of the drone seen from the camera. Then, the three-dimensional relative position of the drone is estimated based on the camera internal parameters, the bounding box information, and the depth map taken by the stereo vision. In addition, the Kalman filter is employed to estimate the position of the drone continuously. Finally, the position estimation performance of the proposed algorithm is evaluated through the experiments.