• Annual Conference of KIPS
• /
• 2023.05a
• /
• pp.667-669
• /
• 2023

데이터의 균형은 객체 인식 분야에서 영향을 미치는 요인 중 하나이다. 본 논문에서는 폐기물 데이터 균형을 위해 Chat-GPT와 Diffusion model 기반 데이터 생성 모델을 제안한다. Chat-GPT를 사용하여 폐기물의 속성에 해당하는 단어를 생성하도록 질문하고, 생성된 단어는 인코더를 통해 벡터화시킨다. 이 중 폐기물과 관련 없는 단어를 삭제 후, 남은 단어들을 결합하는 전처리 과정을 거친다. 결합한 벡터는 디코더를 통해 텍스트 데이터로 변환 후, Stable Diffusion model에 입력되어 텍스트와 상응하는 폐기물 데이터를 생성한다. 이 데이터는 AI Hub의 공공 데이터를 활용하며, 객체 인식 모델인 YOLOv5로 학습해 F1-score와 mAP로 평가한다.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2023.07a
• /
• pp.11-12
• /
• 2023

본 논문에서는 상점 내 캡티브 포털을 활용하여 수집된 주문 정보 데이터를 바탕으로 사용자가 선호하는 메뉴를 추천하는 시스템을 제안한다. 이 시스템은 식품 관련 공공 데이터셋으로 학습된 단어 임베딩 모델(Word Embedding Model)로 메뉴명을 벡터화하여 그와 유사한 벡터를 가지는 메뉴를 추천한다. 이 기법은 캡티브 포털에서 수집되는 데이터 특성상 사용자의 개인정보가 비식별화 되고 선택 항목에 대한 정보도 제한되므로 기존의 단어 임베딩 모델을 추천 시스템에 적용하는 경우에 비해 유리하다. 본 논문에서는 실제 동일한 시스템을 사용하는 상점들의 구매 기록 데이터를 활용한 검증 데이터를 확보하여 제안된 추천 시스템이 Precision@k(k=3) 구매 예측에 유의미함을 보인다.

• KIISE Transactions on Computing Practices
• /
• v.23 no.4
• /
• pp.226-231
• /
• 2017

Word representation, an important area in natural language processing(NLP) used machine learning, is a method that represents a word not by text but by distinguishable symbol. Existing word embedding employed a large number of corpora to ensure that words are positioned nearby within text. However corpus-based word embedding needs several corpora because of the frequency of word occurrence and increased number of words. In this paper word embedding is done using dictionary definitions and semantic relationship information(hypernyms and antonyms). Words are trained using the feature mirror model(FMM), a modified Skip-Gram(Word2Vec). Sense similar words have similar vector. Furthermore, it was possible to distinguish vectors of antonym words.

• The Transactions of the Korea Information Processing Society
• /
• v.3 no.6
• /
• pp.1646-1657
• /
• 1996

This paper presents a color transformation method based on a uniform color image model. Firstly, color variation factors are grouped into identical (multiplicative) factor and independent(additive) one for the color model, and they are modelled by the Gaussian function. The shape of a color cluster in (R, G, B) feature space is an ellipsoid whose elongated major axis correspond to the direction of mean vector. Secondly, the transformation of a color cluster using the model is studied. A transformation method for three dimensional coordinated is described. The proposed method is applied to artificial and natural color images. By the result of experiments, the elongated major axis of each cluster making up the transformed color image aggress with the direction of its mean vector.

• Journal of the Korea Computer Graphics Society
• /
• v.26 no.5
• /
• pp.15-24
• /
• 2020

We present a supervised learning method that estimates the simulation parameters required to simulate the fabric from the static drape shape of a given fabric sample. The static drape shape was inspired by Cusick's drape, which is used in the apparel industry to classify fabrics according to their mechanical properties. The input vector of the training model consists of the feature vector extracted from the static drape and the density value of a fabric specimen. The output vector consists of six simulation parameters that have a significant influence on deriving the corresponding drape result. To generate a plausible and unbiased training data set, we first collect simulation parameters for 400 knit fabrics and generate a Gaussian Mixed Model (GMM) generation model from them. Next, a large number of simulation parameters are randomly sampled from the GMM model, and cloth simulation is performed for each sampled simulation parameter to create a virtual static drape. The generated training data is fitted with a log-linear regression model. To evaluate our method, we check the accuracy of the training results with a test data set and compare the visual similarity of the simulated drapes.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.47 no.6
• /
• pp.75-85
• /
• 2010

This paper presents a robust 3D model hashing based on object feature vector for 3D content authentication. The proposed 3D model hashing selects the feature objects with highest area in a 3D model with various objects and groups the distances of the normalized vertices in the feature objects. Then we permute groups in each objects by using a permutation key and generate the final binary hash through the binary process with the group coefficients and a random key. Therefore, the hash robustness can be improved by the group coefficient from the distance distribution of vertices in each object group and th hash uniqueness can be improved by the binary process with a permutation key and a random key. From experimental results, we verified that the proposed hashing has both the robustness against various mesh and geometric editing and the uniqueness.

• Journal of Energy Engineering
• /
• v.23 no.4
• /
• pp.263-271
• /
• 2014

Corrective thermal performance analysis is required for thermal power plants to determine performance status of turbine cycle. We developed classification method for main feed water flow to make precise correction for performance analysis based on ASME (American Society of Mechanical Engineers) PTC (Performance Test Code). The classification is based on feature identification of status of main water flow. Also we developed predictive algorithms for corrected main feed-water through Support Vector Machine (SVM) Model for each classified feature area. The results was compared to estimations using Neural Network(NN) and Kernel Regression(KR). The feature classification and predictive model of main feed-water flow provides more practical methods for corrective thermal performance analysis of turbine cycle.

• The Journal of the Acoustical Society of Korea
• /
• v.21 no.5
• /
• pp.455-461
• /
• 2002

This paper presents a method that improves the performance of the personal voice dialling system in which speaker independent phoneme HMM's are used. Since the speaker independent phoneme HMM based voice dialing system uses only the phone transcription of the input sentence, the storage space could be reduced greatly. However, the performance of the system is worse than that of the system which uses the speaker dependent models due to the phone recognition errors generated when the speaker independent models are used. In order to solve this problem, a new method that jointly estimates transformation vectors for the speaker adaptation and transcriptions from training utterances is presented. The biases and transcriptions are estimated iteratively from the training data of each user with maximum likelihood approach to the stochastic matching using speaker-independent phone models. Experimental result shows that the proposed method is superior to the conventional method which used transcriptions only.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.17 no.5
• /
• pp.364-371
• /
• 2016

We developed a prediction model of assistance force for the opening/closing of an automobile door depending on the condition of the parking ground. The candidates of the learning models for the operating assistance force were compared to determine the proper force according to the slope and user's force, etc. The reduced experimental model was developed to obtain learning data for the estimation model. The learning algorithm was composed to predict the assistance force to incorporate real assistance force data. Among these algorithms, an Artificial Neural Network (ANN) and Support Vector Machine(SVM) were applied and the adaptability was compared between these models. The SVM provided more adaptability for the learning process of the door assistance force prediction. This paper proposes a system for determining the assistance force to control a door motor to compensate for the deviation of required door force in the slope condition, as needed in the plane condition.

• Journal of Digital Convergence
• /
• v.16 no.12
• /
• pp.301-307
• /
• 2018

The most of the IR focus on the method for searching the document, so the keyword-based IR system is not able to reflect the feature information of the image. In order to overcome these limitations, we have developed a system that can search similar images based on the vector information of images, and it can search for similar images based on sketches. The proposed system uses the GAN to up sample the sketch to the image level, convert the image to the vector through the CNN, and then retrieve the similar image using the vector space model. The model was learned using fashion image and the image retrieval system was developed. As a result, the result is showed meaningful performance.