Earticle

This article is two-fold. The ultimate goal of this article is to provide a big data analysis of 330 reviews of the movie Noryang and to evaluate the Naive Bayes model, the Random Forests model, the DNN model, and the LSTM model in machine learning and deep learning. A point to note is that the name Yi, Sun-shin was the most widely used by viewers, followed by the word movie, and the word general, in that order. A major point of this article is that the name Yi, Sun-shin and the word movie showed up twice as the first keyword. This in turn implies that these keywords are the most noteworthy ones. The sentiment analysis argues that about 75% of viewers think of the film as well-made and that they were highly satisfied with it. In this paper, we used the Naive Bayes model, the Random Forests model, the DNN model, and the LSTM model and made them predict whether each review is positive or negative. The Random Forests model works well for our data, whereas the Naive Bayes model does not. When learning took place 25 times, the DNN model worked well for our data (its accuracy rate is 82.76%). When it comes to the LSTM model, its accuracy did not improve even though learning took place 9 times. Yet, the LSTM model is slightly better than the DNN model with respect to the accuracy rate of test data.

Timber harvesting in forested areas involves the simultaneous operation of heavy forest machines and manual workers, creating a high risk of collision-related accidents. To address these safety concerns, this study developed a deep learning-based human detection sensor system designed for installation on forest forwarders. The system integrates a Raspberry Pi 4 with a Pi-camera running a MobileNetV2-based detection model, optimized for real-time inference under low-power embedded conditions. In addition, ultrasonic sensors were incorporated to measure distances to detected person, enabling accurate localization around the machine. Model training utilized a filtered COCO dataset and achieved optimal performance through augmentation strategies, with the customized M3 configuration and our own test dataset reaching a mean average precision (mAP) of 0.71, precision of 0.95, and recall of 0.99. Experimental evaluations confirmed that the system successfully detected person across various postures, positions, and environmental conditions, with localization errors maintained within acceptable limits. Outdoor tests further demonstrated robust performance even under partial occlusion, although occasional false negatives in complex or low-light scenarios highlighted the need for dataset expansion and sensor fusion. The developed system transmits integrated detection and localization data via CAN bus, confirming its feasibility for deployment in actual forest forwarders. These findings suggest that the proposed sensor system offers a promising solution for enhancing worker safety in mechanized forestry operations and provides a foundation for future smart and autonomous forest machines.

In this study, we consider application of deep learning methods in the cryptanalysis of tiny DES algorithm, which is a DES-like cipher. We develop two types of deep learning architectures to perform the cryptanalysis of tiny DES. It is a known-plaintext attack where the deep learning models only need ciphertext and plaintext pair as training and the learning target is to predict correct plaintext when a ciphertext is given. Simulation results have shown that deep learning methods cannot 100% recover the plaintext of tiny DES but can greatly reduce the analysis difficulty for plaintext recovery.

본 연구에서는 게임 영상과 같은 생성된 영상으로부터 물체를 인식하는 심층 학습 기반 모델의 성능을 향상 시키는 방법을 제시한다. 특히, 실제 영상으로 훈련된 물체 인식 모델에 대해서 게임 영상으로 추가 훈련을 수 행함으로써 물체 인식 성능이 향상됨을 검증한다. 본 연구에서는 심층 학습 기반의 물체 인식 모델들 중에서 가장 널리 사용되는 YoloV2 모델을 이용한다. 이 모델에 대해서 8 종류의 다양한 게임에서 샘플링한 160장의 게임 영상을 적용해서 물체 인식 모델을 다시 훈련하고, IoU와 정확도를 측정해서 본 연구에서 주장하는 게임 영상을 이용한 훈련이 효과적임을 입증한다.

We present a framework that improves the performance of deep learning-based object detection model for generated images including game scenes. In particular, we aim to verify that the additional training using images sampled from game scenes can improve the performance of the object detection model, which was pre-trained using photographs. Among the various object detection schemes including Yolo V1, Yolo V2 and SSD, we employ YoloV2 model, which is one of the most widely used deep learning-based object detection model. YoloV2 model is pretrained using diverse photographs. This model is further trained through 160 game scene images sampled from eight different kinds of games. We select the games that range from realistic scenes and highly deformed scenes. We measure IoU (intersection over union) and accuracy using this model. The comparison between our re-trained model and the original model demonstrates the effectiveness of our strategy.

In the resting state, EEG and fMRI have functional correlations in the low-frequency band, and integrating the two modalities can provide a more comprehensive understanding of brain activity. However, multimodal imaging faces challenges such as high cost and complexity of data fusion. In this study, we developed a Transformer-CNN to generate fMRI data from EEG signals and introduced spatial normalization to compensate for differences in brain structures between sub-jects. Our results showed that the brain structures were normalized to the same extent, so that the model could focus only on predicting the signal values of fMRI, and compared with actual fMRI scans, we obtained PSNR of 25.92 and SSIM of 0.56, which were quantitatively and qualitatively evaluated. Although there are some qualitative limitations for medical device utilization, our ap-proach opens new avenues in neuroscience, especially in environments where simultaneous EEG-fMRI acquisition is not possible. This study highlights the potential of deep learning in advancing multimodal imaging and provides enhanced insights into brain function.

산사태는 인명과 사회 경제적 인프라에 심각한 피해를 초래하는 대표적인 자연재해로, 위험지역의 정밀 한 공간 분할 및 조기 탐지가 재난관리에 필수적이다. 본 연구에서는 U-Net 딥러닝 아키텍처를 기반으로 패치 크기(128×128, 256×256)에 따른 산사태 위험지역 자동 분류 성능을 비교·분석하였다. 연구 대상지는 경북 예천군 보곡면의 급경사지로, 드론 영상 및 포인트 클라우드 자료를 활용해 0.1m 해상도의 디지털 표고 모델(DEM)을 구축하고, 이를 딥러닝 학습 및 평가에 적용하였다. 분석 결과, 128×128 패치 크기 모델은 전체 정확도 70.5%의 정확도를 기록했으며, 비위험과 위험 클래스에서 F1-score 0.582, 0.599를 보여 균형 잡힌 분류 성능을 나타냈다. 반면 256×256 패치 모델은 65.96%의 정확도와 매우 낮은 0.248, 0.007의 F1-score를 기록했으며, NoData 클래스에서는 0.786의 높은 F1-score를 보여 대부분의 예측이 NoData 영역에 집중됨을 나타냈다. 이러한 결과는 작은 패치 크기가 지역적 지형 변동성을 포착하고 산사태 감지 성능을 향상시키는데 더 효과적임을 시사한다.

Landslides are natural disasters that cause severe damage to human life and socio-economic infrastructure, making precise spatial delineation and early detection of hazardous areas essential for disaster management. This study compares automated classification performance of landslide hazard zones using U-Net deep learning architecture with different patch sizes (128×128 and 256×256). The study area is a steep slope in Bogok-ri, Hyoja-myeon, Yecheon-gun, Gyeongbuk Province, where a high-resolution DEM (0.1 m) was constructed using drone imagery and point cloud data. The 128×128 patch model achieved 70.5% accuracy with F1-scores of 0.582 and 0.599 for non-hazard and hazard classes, respectively, indicating balanced performance. In contrast, the 256×256 patch model yielded 65.96% accuracy, much lower F1-scores of 0.248 and 0.007, and a higher NoData F1-score of 0.786, reflecting predictions focused on NoData zones over hazard areas. Results suggest smaller patch sizes better capture local terrain variability and enhance landslide detection performance.

A specific area of research in artificial intelligence, known as deep learning (DL) has turned into a strong source for the solutions of complicated issues in computer vision and many more. A real application of that is real and fake faces detection. Detection of is real and fake faces became increasingly important these days with increasing deepfake technology. Fake images pose great dangers to information security, trustworthiness in multimedia content, and even to society's stability. In his proposal the design of a deep learning-based model in VGG-16 architecture to make high accuracy and reliability distinctions between real and fake faces. The performance of the proposed model was evaluated by a number of metrics, including accuracy, specificity, recall, precision, and misclassification rate. The results showed that the model obtained an excellent accuracy of 99.61% with a very low misclassification rate of 0.39%. It obtained perfect specificity of 99.15%, which means all fake faces were identified correctly, and a precision value of 99.29%, ensuring that all faces classified as real were indeed real. The recall of the model was high, at 100%, meaning nearly all real faces were correctly identified. The obtained results are the proof of how effective DL and, in this case, using a pre-trained model like VGG-16, is at recognizing real and fake faces. It shows how strong and reliable the proposed.

최근 인공지능의 딥러닝과 머신러닝을 이용한 예측시스템에 관한 연구가 활발히 진행되고 있다. 인공지능의 발전으로 인해 투자관리자의 역할을 인공지능을 대신하고 있으며, 투자관리자보다 높은 수익률로 인해 점차 인공지능 으로 거래를 하는 알고리즘 거래가 보편화하고 있다. 알고리즘 매매는 인간의 감정을 배제하고 조건에 따라 기계적으로 매매를 진행하기 때문에 장기적으로 접근했을 때 인간의 매매 수익률보다 높게 나온다. 인공지능의 딥러닝 기법은 과거 의 시계열 데이터를 학습하고 미래를 예측하여 인간처럼 학습하게 되고, 변화하는 전략에 대응할 수 있어 활용도가 증가하고 있다. 특히 LSTM기법은 과거의 데이터 일부를 기억하거나 잊어버리는 형태로 최근의 데이터의 비중으로 높 여 미래 예측에 사용하고 있다. 최근 facebook에서 개발한 인공지능 알고리즘인 fbprophet은 높은 예측 정확도를 자랑하며 주가나 암호화폐 시세 예측에 사용되고 있다. 따라서 본 연구는 fbprophet을 활용하여 실제 값과 차이를 분석하고 정확한 예측을 위한 조건들을 제시하여 암호화폐 자동매매를 하기 위한 새로운 알고리즘을 제공하여 건전한 투자 문화를 정착시키는 데 이바지하고자 한다.

Recently, research on predictive systems using deep learning and machine learning of artificial intelligence is being actively conducted. Due to the development of artificial intelligence, the role of the investment manager is being replaced by artificial intelligence, and due to the higher rate of return than the investment manager, algorithmic trading using artificial intelligence is becoming more common. Algorithmic trading excludes human emotions and trades mechanically according to conditions, so it comes out higher than human trading yields when approached in the long term. The deep learning technique of artificial intelligence learns past time series data and predicts the future, so it learns like a human and can respond to changing strategies. In particular, the LSTM technique is used to predict the future by increasing the weight of recent data by remembering or forgetting part of past data. fbprophet, an artificial intelligence algorithm recently developed by Facebook, boasts high prediction accuracy and is used to predict stock prices and cryptocurrency prices. Therefore, this study intends to establish a sound investment culture by providing a new algorithm for automatic cryptocurrency trading by analyzing the actual value and difference using fbprophet and presenting conditions for accurate prediction.

Masonry is a common type of construction that uses mortar to bind individual units, such as brick or building stones, together to construct the structure. Even though masonry structures are durable, multiple factors such as the quality of mortar, workmanship, and harsh environment could greatly reduce the structural integrity, leading to defects and even human loss. Thus, it is crucial to perform the maintenance process regularly. Previously, the maintenance relied mainly on inspectors, who inspected the masonry structures to find cracks and determine the seriousness. However, this process is error-prone, costly, and time-consuming. As a result, this study proposes a fully automated masonry crack segmentation framework that robustly identifies various types of masonry cracks. In addition, the length of the segmentation cracks, which has been ignored in previous studies, is also computed.

This research aimed to assess the possibility of detecting forest degradation using time-series satellite imagery and three different deep learning-based change detection techniques. The dataset used for the deep learning models was composed of two sets, one based on surface reflectance (SR) spectral information from satellite imagery, combined with Texture Information (GLCM; Gray-Level Co-occurrence Matrix) and terrain information. The deep learning models employed for land cover change detection included image differencing using the Unet semantic segmentation model, multi-encoder Unet model, and multi-encoder Unet++ model. The study found that there was no significant difference in accuracy between the deep learning models for forest degradation detection. Both training and validation accuracies were approximately 89% and 92%, respectively. Among the three deep learning models, the multi-encoder Unet model showed the most efficient analysis time and comparable accuracy. Moreover, models that incorporated both texture and gradient information in addition to spectral information were found to have a higher classification accuracy compared to models that used only spectral information. Overall, the accuracy of forest degradation extraction was outstanding, achieving 98%.

Lossy image compression provides an efficient solution to the exchange and storage of image data for consumer applications. The design of lossy algorithms is based on a principle to discard information that are not perceivable by human visual system (HVS). With the popularity of deep learning models (DL) in computer vision (CV), it is necessary to characterize the loss in image quality with respect to computer vision systems as well. Recent studies have analyzed the image distortions resulted from blur and noise, mainly from an adversarial attack perspective. However, fewer studies have dealt with the lossy nature of the JPEG algorithm. Therefore, the current study presents a quantitative assessment of different types of data loss that occurs due to chroma subsampling, quantization, and rounding functions of the JPEG algorithm. In addition, we have analyzed impact of different interpolation methods that are used for chroma upsampling. The analysis have shown that for compression savings, performing either subsampling or quantization preserved the model accuracy while their combination degraded the accuracy by 6%.

The main visual feature is surface cracks, which are caused by loopholes that are embedded into the structures due to manufacturing faults and overloading factors. The structural health monitoring must be precise and more efficient for detecting surface cracks in concrete structures. Human inspection is used to identify damage on concrete surfaces. But, these traditional visual observation techniques are not more effective for large concrete structures. Moreover, this outmoded human labor practice for crack detection is intensive, expensive, and inefficient. To predict potentially hazardous situations caused by cracks on concrete surfaces, it is crucial to have an efficient, fast, and well-organized inspection system for concrete surface cracks. The automatic crack detection system must be effective in identifying cracks, damage, and segmenting them. In this research, a deep learning (DL) algorithm model is active for crack detection in concrete structure images to assess the influence on structural health. This design work is projected to provide a fast and active solution for identifying dust/duct type to prevent power losses using an image classification model based on DL. The VGG-16 DL model significantly analyzes the precision and accuracy of identifying the crack surfaces on concrete structures. The exceptional performance of the projected model achieves a training accuracy of 99.99% and a test accuracy of 99.96%, with an F1 score of 0.9995, precision of 0.9995, and a sensitivity of 0.9997. This is more precise, costeffective, and more efficient than human resources to find the defect on concrete surfaces that may support the healthy life of concrete structures.

본 논문은 정신분석학과 인지심리학의 무의식 개념과 딥러닝의 내부 작동 방식 사이에 존재하는 구조적, 개념적 유사성을 탐색하고, 이를 통해 인공지능의 작동 원리를 인문학적 관점에서 새롭게 재해석하고자 하는 것이다. 심 리학 이론에서 인간의 정보처리는 의식적 영역보다도 자동적이고 암묵적인 무의식에 더 큰 영향을 받으며 작동한 다고 본다. 본 논문은 AI 의 대표적 실현 방법인 딥러닝 모델의 신경망 내부 작동 방식이 감추어진 잠재 공간을 생성한다는 점에서 일종의 “기계적 무의식” 이라 보고 논의를 진행하였다. 구체적으로 신경망의 비선형적 처리, 층간 억압 구조로 만들어 지는 잠재 공간 개념을 인지심리학과 정신분석학의 무의식, 집단무의식, 동시성 이론으 로 비교함으로써, AI 작동 메커니즘이 인간 심리 구조의 은유로서 이해될 수 있는 가능성을 제시하였다.

This paper explores the structural and conceptual parallels between the notion of the unconscious in psychoanalysis and cognitive psychology, and the internal operational mechanisms of deep learning models. Through this comparison, the study aims to reinterpret the functioning of artificial intelligence from a humanities- oriented perspective. Psychological theories posit that human information processing is influenced more strongly by automatic, implicit, and unconscious mechanisms than by conscious deliberation. Building on this view, the present work conceptualizes the hidden internal processes of deep learning&# 8212;particularly the formation of latent spaces generated through nonlinear computations and inter- layer suppression—as a form of “mechanical unconscious.” By comparing these latent computational structures with the psychological constructs of the unconscious, the collective unconscious, and synchronicity proposed in psychoanalysis and cognitive psychology, this paper suggests that the operational mechanisms of AI can be meaningfully understood as metaphors for human psychological processes.

본 연구는 딥러닝 기반 자연어처리 기법을 활용하여 한국과 중국의 소셜미디어 공간에서 나타나는 반일 담론의 정서적･의미적 구조를 비교 분석하였다. 트위터(X), 웨이보, 바이두에 서 수집된 약 9,000건의 온라인 게시글을 바탕으로, 워드임베딩, t–SNE 시각화, 감성어 사전 기반 감정 분석 등의 기법을 적용하여 한･중 반일 감정의 언어적 표출 방식을 실증적으로 고 찰한 것이다. 분석 결과, 한국과 중국 모두 반일 정서가 강하게 나타났으나, 그 표현 방식과 정서적 구현 양상에서 상이한 특성이 관찰되었다. 한국의 반일 담론은 역사적 피해 의식을 바탕으로 정치 적 이슈, 소비자운동, 대중문화와 같은 다양한 분야로 확산되어 있으며, 부정적 감정과 일상 적 수용 정서가 혼재된 양가적인 구조를 형성하고 있는 것으로 나타났다. 특히 ‘불매’, ‘역사 왜곡’, ‘우익’ 등과 같은 키워드는 정서적으로 강한 반감과 함께, 국가 정체성 및 도덕적 정당 성을 강조하는 담론 구조와 연결되어 있었다. 반면 중국의 반일 담론은 난징대학살, 식민지배, 전쟁 범죄 등 역사적 피해 경험에 대한 집 단 기억을 중심으로 구성되어 있었으며, ‘仇恨(증오)’, ‘侵略(침략)’, ‘右翼(우익)’ 등과 같은 키 워드를 통해 민족주의적 감정과 도덕적 비판이 강하게 표출되고 있었다. 동시에 일본 대중문 화 콘텐츠에 대한 관심과 소비 행태에 있어서는 별도의 정서적 층위가 형성되고 있어, 반감 과 수용이 공존하는 이중적인 양상이 확인되었다. 감정어 분석에서도 한국은 ‘역사’, ‘사과’, ‘자유’, ‘친일파’ 등의 단어가 긍정･부정 감성어 와 복합적으로 연결되며, 반일 정서가 단순한 외부 비판을 넘어 국내 정치적･도덕적 질서와 도 연결되어 있는 양상을 보였다. 중국 역시 부정 감성어를 중심으로 ‘전범’, ‘군국주의’, ‘역 사 수정’ 등의 키워드가 다수 확인되었으며, 이는 일본에 대한 도덕적 불신과 정치적 반감이 여실히 드러났다. 이러한 결과는 한국과 중국의 반일 감정이 유사한 역사적 배경을 공유하면서도, 각국의 사 회문화적 맥락과 담론 형성 방식에 따라 상이한 양상으로 전개되고 있다는 것을 확인해 주는 것이다. 나아가 디지털 공간에서의 집단 정서가 단순한 감정 표출을 뛰어넘어, 역사 인식과 문화적 태도의 차이를 반영하는 중요한 지표로 활용될 수 있다는 것을 보여준 것이다.

This study applied deep learning–based natural language processing methods to analyze anti–Japanese discourse in Korean and Chinese online spaces. Using approximately 9,000 posts collected from Twitter (X), Weibo, and Baidu, the research employed word embedding, t–SNE visualization, and sentiment lexicon–based analysis to investigate how anti–Japanese sentiment is linguistically structured in each country. The findings reveal that while both Korea and China exhibit strong anti–Japanese sentiment, the ways in which these emotions are expressed differ significantly. In Korea, anti–Japanese discourse expands beyond historical grievances to include political issues, consumer activism, and cultural preferences, forming an ambivalent emotional structure. Keywords such as “boycott” and “right–wing” reflect a discourse that is intertwined with national identity and domestic moral narratives. China’s discourse, on the other hand, is more deeply rooted in historical trauma and nationalist emotion, centering on themes such as invasion and war crimes. At the same time, admiration for Japanese pop culture exists in a separate emotional layer, reflecting a coexistence of hostility and cultural appreciation. These results suggest that anti–Japanese sentiment in both countries is shaped by distinct sociocultural contexts, and that online discourse functions not only as a medium of public opinion but also as a space where historical memory and national identity are actively constructed.

PKD(다낭성 신장 질환)는 신장에 낭종이 형성되어 치명적인 합병증을 유발할 수 있는 질환으로, 완치가 불가능하여 정확한 진단과 약물을 이용한 평생 관리가 필요하다. 그러나 환자 별 다양한 요 인으로 인해 진단이 복잡하여 신부전으로 진행될 위험이 높다. 본 연구에서는 다기관에서 수집한 3D MR 이미지와 임상 데이터를 활용하여, 이미지에서 ResNet-152 모델을, 임상 데이터에서 MLP 를 사용해 특징을 추출하고, 이를 결합한 멀티모달 분류 방법을 제안했다. 제안된 방법은 정확도 73.9%, F1-score 0.754를 달성하였다.

본 연구는 스마트팜 자동화를 위한 딥러닝 기반 엽채류 이미지 분류 시스템을 개발하고, 그 성능을 심 도 있게 분석하였다. 기본적인 3층 구조의 CNN 모델뿐만 아니라 ResNet, Inception과 같은 고도화된 딥러닝 모델을 활용하여 엽채류 이미지를 분류하였으며, 각 모델의 성능을 비교하였다. 연구에 사용된 데이터는 스마 트팜 환경에서 촬영된 고해상도 이미지와 공개 데이터셋으로 구성되었으며, 다양한 전처리 및 데이터 증강 기 법을 통해 모델의 학습 효과를 극대화하였다. 그 결과, 딥러닝 기반 분류 시스템은 기존의 원본을 기반으로한 딥러닝에 비해 더 높은 정확도와 효율성을 보여주었으며, 스마트팜의 자동화를 통한 생산성 향상과 비용 절감 에 기여할 수 있음을 확인하였다. 이러한 연구는 스마트팜 기술의 발전에 중요한 역할을 할 것이며, 농업 자동 화의 미래를 위한 기술적 토대를 마련하는 데 기여할 것으로 기대된다.

This study developed a deep learning-based leafy vegetable image classification system aimed at smart farm automation and conducted an in-depth performance analysis. In addition to the basic 3-layer CNN model, advanced deep learning models such as ResNet and Inception were employed to classify leafy vegetable images, and the performance of each model was compared. The dataset used for the study consisted of high-resolution images captured in a smart farm environment, along with publicly available datasets. Various preprocessing and data augmentation techniques were applied to maximize the effectiveness of model training. As a result, the deep learning-based classification system demonstrated higher accuracy and efficiency compared to traditional deep learning approaches based on original data. It was confirmed that the system could contribute to improving productivity and reducing costs through smart farm automation. This research is expected to play a key role in advancing smart farm technologies and contribute to laying the technological foundation for the future of agricultural automation.

The use of drones is rapidly increasing in sports, photography, and entertainment purposes because of their affordable price and lightweight nature. However, this potential increment in the of drone of drone is creating safety and security threats. The detection of drones is necessary to overcome these issues. The detection of drones may be challenging because of the presence of other aerial objects like aircraft and birds. The existing systems used for drone detection employed a small dataset with a lack of diverse images. To overcome the limitation in previous studies, in this study, we used a largescale dataset drone images dataset. We conducted experiments on different You Only Look Once Version 8 (Yolov8) models using this dataset. All the trained models are evaluated in terms of precision, recall, mAP50, and mAP50-95. Yolov8x exhibits high performance in terms of precision, recall, mAP50, mAP50-95 models among other models which shows the superiority of Yolov8x in drone detection technology.

인공지능 기반 반려동물 피부영상 진단기술은 전세계적으로 아직 초기단계에 머물러 있다. 특히 풍성한 털로 덮인 반려동물의 피부영상에서 털을 제거하는 것은 해결되지 않은 기술적 난제로 남아 있다. 본 논문은 반려동물 고배율 피부영상에서 털 제거를 위한 필수적인 전처리 단계로서 털 영역을 분할하기 위해 딥러닝 모델인 U-net과 LadderNet의 적용 가능성을 시험하고 기존의 영상처리 기반 털 영역분할 알고리즘과 성능을 비교 분석한다 (Ronneberger et al. 2015, Zhuang 2018). 반려동물 피부영상 뿐만 아니라 사람 피부영상, 가짜털을 생성한 합성영상을 이용하여 딥러닝 모델을 훈련한 효과도 분석하였다. 실험 결과, 반려동물 피부영상 데이터셋을 사용하여 훈련된 LadderNet이 반려동물 피부영상에 대하여 기존 영상처리 기반 방법 및 U-net보다 더 높은 F1-score, 정 확도 및 AUC 값을 보였다. 반면 합성영상을 사용한 모델 훈련은 실제 피부영상에 대해서는 성능 개선 효과가 없었 지만 합성영상에 대해서는 좋은 성능을 보였다.

Artificial intelligence technologies for pet skin image diagnosis is still in its infancy. Especially, it remains unsolved to remove abundant hairs from pet skin images. In this paper, we evaluated the applicability of two deep learning models, including U-net and LadderNet, for hair segmentation as an essential preprocessing step for hair removal in pet microscopic skin images, and we compared their performance with an existing method based on image processing. In particular, those models were trained not only with pet and human skin images but also with synthetic skin images of fake hairs. In experimental results, the performance of LadderNet for hair segmentation in pet skin images, measured as F1-score, accuracy and AUC, was superior to U-net as well as the image processing method when it was trained with the pet skin image dataset. On the other hand, training both models with synthetic skin images was effective not in real skin images but in synthetic skin images.