We have been applied sparse modeling AI to identify factors and predict in power demand. In former study, we only used data in summer season. By contrast former study, we use yearly data to identify factors and predict in that. Sparse modeling AI shows that factors of power demand are different data in summer season and yearly data. In addition to that, sparse modeling AI succeeded in achieving about 5% averaged absolute prediction error.

In Energy-from-Waste plants, combustion is dependent on many complex factors and a sign of abnormal states appears in other sensors. We have proposed fuzzy relational maps of many sensors (FuRMS) for expressing relationships between many sensors as relations between fuzzy sets of sensor values. We constructed several fuzzy relational maps for combustion states (a stable state and three abnormal states) using the data set to be their states after a certain time. With evaluating unknown data for the maps, we predict the combustion state after a certain time, but cannot do it very well when combustion state transfers to an abnormal state, i.e., a certain sensor value exceeds a threshold one. In this paper, we separate the data set into two data sets, to transfer to the other state and to remain the same states after a certain time. We construct several fuzzy relational maps using the data set to transfer to the other state, to predict the combustion state after a certain time with the maps. The result shows that the maps using the separated data can predict with higher accuracy than the unseparated data.

In this paper, we improve the iterative rule generation mechanism for fuzzy modelling. First,to get a more accurate model, an precedent narrowing algorithm is proposed. Next, a random sampling method is also introduced to obtain errors of each candidate models to make modelling faster. According to the results of simulations, these methods work effectively enough.

We investigated the effectiveness of an interactive artificial bee colony (IABC) method to retrieve multimodal candidate solutions for real users. The IABC method retrieves various candidate solutions using three types of bees (i.e., employed, onlooker, and scout bees). In this study, we performed experiments to examine the effectiveness of the IABC method for real users using running shoe designs as an evaluation object. The results showed that the IABC method could retrieve more multimodal candidate solutions than the conventional interactive genetic algorithm method.

In this research, we propose an input method using flick input without pointing for the purpose of reducing the burden and fatigue of the conventional method using pointing in Japanese text input on VR. We developed a prototype of the proposed method and compared it with the two types of keyboards developed by the existing method. The proposed method inputs characters using only the stick operation of the controller. The action of tilting the stick is a pseudo flick operation, and the key selection and flick input are used properly with the index finger trigger of the controller. We asked the subjects to play a typing game on each keyboard, and measured input speed and accuracy, usability from questionnaires, and mental load. While the flick-pointing keyboard has the fastest input speed, there were many erroneous inputs. The proposed method had the slowest input speed, but the input accuracy was higher than that of the flick-pointing keyboard. On the other hand, in terms of usability and mental load, the Japanese syllabary keyboard performed best, and the proposed method and the flick pointing keyboard were found to be comparable.

In this paper, we assume that the user’s posture when using a laptop computer affects the keystrokes, and examine the effect of the difference in posture on the keystroke continuation authentication accuracy. Keystroke data when using a laptop computer were collected from 5 subjects for 3 different postures (basic posture and 2 other postures selected by each subject). In the experiment, we compared the accuracy of continuous authentication with and without posture information. We employed two authentication methods, a machine learning-based and a distance-based. The DTM protocol was used to evaluate the accuracy of continuous authentication. As a result, we confirmed that the test data with posture information didn’t significantly affect the continuous authentication accuracy in either the machine learning-base or the distance-base.

This paper proposes a method for matrix factorization-based collaborative filtering (CF) that focuses on feedback to specific items. Although a large amount of user information is usually required for accurate recommendation, privacy concern makes it difficult to collect user information. As a solution, we focus on applying CF to a small amount of feedback on a few items by developing a method for determining the items for which user feedback can be efficiently obtained (Probing Items), and a method for learning a recommendation system focusing on the probing items. This paper focuses on the latter and proposes a method for learning a matrix factorization-based CF, in which probing items are considered at every epoch while other items are randomly dropped. Experimental results show the effectiveness of the proposed method.

This paper proposes dynamic contextual features based on timestamp information for music recommender systems. Recently, with the development of communication technology and audiovisual devices, online music streaming has become more available to users, so it has become especially important to improve the accuracy of music recommender systems in order to enhance the users ’experience. On the other hand, due to the way how music is listened to and the characteristic of music items, it is difficult to obtain explicit feedback such as ratings from listeners. Therefore, context-aware music recommender systems that use the listener’s contextual information as auxiliary information have become one of the popular research directions. However, existing context-aware music recommender systems have mainly focused on listeners’ static contextual information such as nationality and languages. To consider dynamic contextual information, this paper focuses on the timestamps of listening events and proposes new contextual features about sequence of listening events. An experiment is conducted by incorporating the proposed features into FMs (Factorization Machines), of which the result shows the effectiveness of the proposed features.

A particular pattern of human-to-human touch reduces stress of a human and makes a human’s feeling comfortable. The human-to-human touch is used for therapy of psychological disorders recently. Although it is known that hand temperature, touch speed, and visual information affect the comfort of the touched person in human-to-human touch, it is not clear how the magnitude of these parameters affects the comfort of the touched person. The final goal of this study is to clarify the combination of the magnitude of these parameters which maximize the comfort of touched parson. As a fundamental study of this final goal,we examined the effectiveness of a method for measuring changes in human emotion induced by touch using changes in pupil diameter.

The occurrence that lack of entrainment may lead negatively impact conversational success, therefore the final purpose of our study is to use robotic assistance to increase entrainment occurs in human-human interaction. In this paper, we processed the data detected from the dry electrode type of brainwave sensor by two kinds of entrainment assessment approaches calculated by Pearson Correlation Coefficient (PCC) and Dynamic Time Warping (DTW). The result showed DTW algorithm is more suitable to assess the entrainment phenomenon than PCC, because the phenomenon may have different lengths of time-series data and delays, such as brainwaves entrainment.

We previously proposed a method for estimating missing values in a table. In this method, certain rules with one condition are generated at first, then we assign a value to each missing value so as to generate a certain rule with the highest accuracy value as many as possible. In this paper, we propose another method for estimating the missing value in a table. This method is more realistic than the previous method.

The family of possible coverings and a family of possible indiscernibility relations have a lattice structure with the minimum and maximum elements. This is true for the family of possible maximal descriptions, but is not for the family of possible minimal descriptions and the family of possible sets of close friends. As was shown by Lipski, what we can obtain from an information table with incomplete information is the lower and upper bounds of information granules. Using only two coverings: the minimum and maximum possible ones, we obtain the lower and upper bounds of lower and upper approximations. There, therefore, exists no difficulty of the computational complexity in our approach.

Annotated evidential logic has been studied as a basic theory for dealing with the evidence of propositions and the degree of certainty of beliefs. In this paper, we examine to compare the interpretation of truth-value of propositions in annotated evidential logic with the interpretation of uncertainty by the approximation of variable precision rough set. Using a characteristic function based on the interpretation of a rough set extended to four-value logic, we give truth-value interpretation considering uncertainty for propositions. We also show that a deduction system of annotated evidential logic based on rough sets and discuss an application for annotated evidential logic with semantics interpretation with rough sets.

With the advent of an ultra-smart society, it is desirable to have a mobile assistive robot that can act flexibly like a human without requiring prior information or trial-and-error learning, by appropriately judging the space to be paid attention to and making decisions using only time-series observation information of the surrounding area in public spaces where many people come and go freely. In conventional fuzzy control of mobile robots, the degree of danger was calculated using a membership function based on distance information at the current time for action control, which often failed to take time series into account, and in some cases failed to avoid moving obstacles. In the case of humans, approaches to avoidance are different because they take into account spatio-temporal relationships, such as whether an obstacle is approaching the human or only the human is approaching the obstacle. Therefore, it is thought that flexible behavior is realized by natural look-ahead, such as increasing the weight of the avoidance action in advance and starting the avoidance action, or paying appropriate attention to the approaching object, depending on the ever-changing situation. In order to achieve more flexible obstacle avoidance and target tracing behaviors like humans, we propose a multi-objective behavior coordination based on perceiving the situation and spatial attention-based sensory network, and verify the effectiveness of the proposed system.

This paper is concerned with a mixed control problem of H1 disturbance attenuation control and guaranteed cost control for nonlinear systems, which are modelled by Takagi-Sugeno fuzzy model. We propose a controller that not only attenuates the disturbance but also improves the control performance based on multiple Lyapunov function, which uses the integral of the membership function. Finally, a numerical example is provided to illustrate the validity of the proposed approach.

The number of people who use bicycles in Japan has been increasing rapidly, and many traffic accidents have been occurring. There were 70,000 traffic accidents involving bicycles in 2021, accounting for 22.8% of all traffic accidents in Japan, and this ratio is increasing every year. In this paper, we propose a bicycle collision prediction system using object detection.

Creating cost estimation sheets is one of business affairs in the field of construction. The cost estimation sheets are required to summarize necessary items, quantities, and prices for construction work in process. Since the contents of the cost estimation sheets directly affect the details of works, it must be always accurate without any errors. Therefore, the supervision process based on the past use cases is required to confirm the validity of the sheets in the final step. However, the past cost estimation sheets are stored separately from the other construction projects, so it is necessary to manually check them one by one to find the desired use case, which requires an enormous amount of time and effort. To solve this problem, we prepare a database from the past cost estimation sheets and develop an efficient system that can find the past use cases in the audit process.

The single distance-based toll has been applied to overall urban expressway networks. The route-based toll system for urban expressways is introduced in terms of spatial demand response pricing policy. The combinatorial optimization problem for individual distance-based toll for each route should be formulated. Therefore, the ant colony optimization method is applied as an essential swarm intelligence technique in the study. The practical modification and efficient promotion of ant colony algorithm would be discussed. Finally, the efficient combination of route-based tolls is proposed to reduce the social dead weight loss on the urban networks.

Recently, BRT (Bus Rapid Transit system) has been developed in Japan. In particular, the formation of inhabitant attitudes for public transport along the BRT line corresponding to the alternative routes to existing underground. The questionnaire survey is conducted in social experiment of BRT systems. Fuzzy decision tree with fuzzy decision is applied to summarize the significant factors for BRT service usage. The effective BRT transport policy can be proposed from the attitudes of inhabitants because the decision process for BRT can be described obviously with fuzzy decision tree.

Food waste and food loss is becoming a big social problem in the world. The main reason of throwing away food before being eaten is best-before date expiration. This paper presents a content-based recommender system for meal planning that aims to reduce the amount of expired food in a family. Moreover, such a recommender system is required to not only extract user's preferences but also suggest well-balanced diet. The recommender system therefore has the multi-objective functions to minimize the amount of expired food and adjust caloric intake. Furthermore, we applied a multi-island genetic algorithm to the recommender system and used weighting factors to each objective that users can adjust according to their preference.

This paper presents a virtual reality system with an immersive head-mounted display for assessment of unilateral neglect. Unilateral Spatial Neglect (USN) is a neuropsychological condition in which a deficit in attention to and awareness of one side of space is observed. Paper-and-pencil tests are used as a quick assessment for the diagnosis of USN in rehabilitation field. However, the sensitivity of these tests is not sufficient for evaluating patient's adaptability to more complex and practical tasks. In this study, we therefore developed a virtual reality system to simulate rehabilitation tasks in 3D environment and measure behavioral data such as gaze point, posture and movement. Moreover, we propose a method of feature extraction with topological mapping for measuring the cognitive function on rehabilitation tasks in the virtual space. Furthermore, this paper presents an experimental example to discuss the applicability of the proposed method.

In recent years, research on lip-reading technologies using machine learning and image processing has been conducted in Japan and overseas. Many of these research have focused on cognitive methods and cognitive accuracy, but few have focused on hearing-impaired person who is actually communicating by using lip-reading. We think that even speakers with high cognitive accuracy in lip-reading technologies may not be easy to lip-read for hearing-impaired person. Therefore, in this research, we conducted an experimental survey to numerically elucidate how to speak easily for hearing-impaired person who communicate by using lip-reading on a daily basis.

In this presentation, we report a tentative system of memory retention by motility memory using virtual reality. No matter how much information is stored in the short-time memory, if it is not retained in the brain as a long-term memory, the information will be forgotten with time. Therefore, there is a need for a tool that enables any person to perform memory consolidation easily and more efficiently. Therefore, we focus on motility memory and VR, and investigate whether users can efficiently fix their memories by moving themselves in a virtual space.

We report on a tentative system of communication considering granularity and fuzzies of information in a virtual space. In this study, we develop a system that enables users to visit Chitose University of Science and Technology in a virtual space where they can interact with other users. The system supports communication by promoting self-disclosure and providing advice on conversation, considering the granularity and fuzzies of information, thereby supporting the intimacy of interpersonal relationships.

Information granularity has an important role in human communications. That is, difference of information granularity may interfere in flow of user communication. In this presentation, we make a basic consideration of rough-set-based information granularity in metaverse communications.

In this paper, we study an anomaly detection method for human flow which is significantly different from normal cases due to accidents or natural disasters. Human flow can be assumed to have typical patterns such as going to work and leaving work in the daily life of the people. Developing an anomaly detection method can lead to the discovery of the hidden cause. In this paper, we study a method which aims to detect anomalies in human flow, taking as an example in operation status of railways. We confirmed that our method detects actual suspension of operations.

Intelligent robot system risk management is an urgent issue to be solved, especially in the cybersecurity area. Enormous demands for intelligent and remote mobile robot control are increasing while keeping security. A Cyber Security Challenge was introduced in the 2019 Intelligent Ground Vehicle Competition to accentuate the dire need. The Challenge emphasizes the understanding of RMF (Risk Management Framework) and its actual implementation and demonstration the mobile robot. Since RMF is designed for general purposes, it cannot directly be applied to the mobile robot scenario. This paper develops a Slack bot-based new interactive risk management suggestion system for the mobile robot scenario. By providing mobile robot scenarios to be considered cybersecurity, the Slack bot becomes an interactive questioner to find appropriate recommended risk management action. To confirm the validity of the proposed suggestion system, we decompose various scenarios from our IGVC2019 design reports and compare recommendations.

The basic objective of an autonomous mobile robot is to navigate to its destination while avoiding obstacles. In general, it is advantageous to travel the shortest distance to the destination, but in the real world, there are points that need to be bypassed and courses that can be taken to the destination more safely. Therefore, by assigning waypoints in advance, it is possible to navigate more safely to the destination, and this method is called waypoint navigation. To perform waypoint navigation, a map is required to determine the exact distance and direction to the destination and waypoints, as well as the area where waypoints can be located. However, this method can consume a lot of time and cannot handle unknown environments for which sensor data has not been obtained in advance. In this paper, we describe a novel waypoint navigation system for mobile robots that uses electronic maps and aerial photographs that are easily available in advance. To validate the proposed system, actual electronic maps and aerial photographs were used to generate maps suitable for mobile robot navigation. In addition, a manual and automatic waypoint generation system was developed to generate waypoints for different applications. Based on the generated maps and waypoints, the mobile robot was able to perform waypoint navigation in a real outdoor environment.

Drawing systems for blind people have been developed because it is difficult for them to draw without assistance. Some system has been developed from previous research that enable them to draw using pens and gestures. However, it is difficult for those systems to draw precise figures because they cannot specify detail coordinates when drawing. In addition, some blind people use existing graphic description languages to create precise figures. However, it is a burdensome task for them to think about the drawing process. Based on this background, we are developing an object-oriented graphic description language that enables precise drawing figures and is accessible for blind people. In this paper, we describe the graphic description language we have developed so far.

The phenomenon in which a specific sound gives a specific image to a person is called sound symbolism. There have been many reports on sound symbolism in linguistics, psychology and other fields. However, the reports have not clarified all of sound symbolism. In order to clarify sound symbols, we would like to collect examples of sound symbolism and investigate by bottom-up analysis. Our study took up God’s name, which is considered to be one of the examples of sound symbolism, and investigated the acoustic features of Japanese and foreign God’s name. We used CNN (Convolution Neural Network) and Grad-CAM for the investigation, and analysed the features based on the classification results.

Sound symbolism is known as a phenomenon in which the sound itself evokes a specific impression. Although there have been many reports on sound symbolism, the whole aspect of sound symbolism is still unknown. In order to clarify the whole aspect of sound symbolism, we are attempting to collect and investigate many examples of sound symbolism. In this study, we hired “gunshot onomatopoeia” and “close combat onomatopoeia”, which are considered to be examples of sound symbolism, and investigated the differences in the acoustic features of the two. Machine-learning classifications were used for the investigation, and the differences were analysed from the classification results.

In this study, we analyze impressions of Instagram images of cosmetic brands and examine concepts and image strategies of each brand. Firstly, Instagram images of cosmetic brands are classified, and impression evaluation experiments are conducted for each classified image of each brand. Factor analysis of acquired Kansei data provides the impression image space of cosmetic brands is obtained. Then, we compared the cosmetic images placed in the impression space and the concept of each brand.

Minutes of political meetings of national and local government assemblies are publicly available. Since the meeting minutes contain discussions by several representatives, unlike texts that explain the conclusions, there is a possibility that the process leading to the conclusions can be known. In this paper, we use a large-scale language model to analyze the structure of the minutes and attempt to identify the issues that more than one representative discussed.

In this study, we will investigate the effects of the assignment of speech bubbles and the difference of text presentation (Horizontal, Vertical) on viewers' understanding of details of a video, based on viewers' gaze data. Presentations without speech bubbles or with vertical writing increase the comprehension of contents of the subtitles because the subtitles have been read for a longer duration, but decrease the comprehension of details of the video. Horizontal writing enables viewers to read the subtitles in a shorter time and increases viewers' eye movement. Therefore, it is appropriate to present subtitles horizontally with speech bubbles, taking into account the understanding of details of the video.

The purpose of this study is to assist in the conservation activity minority languages. The activity includes the extraction of words from the collected texts. This study aims to support this task in particular. Unsupervised morphological analysis is used for this purpose, but the conventional method, NPYLM, does not perform well with the small amount of collected texts. In this study, we proposed a method to replace some of the word candidates obtained by NPYLM with single letters and apply NPYLM again. We report its effectiveness.

With the rise of machine learning, it is constantly using pre-trained language models on large-scale data in language processing. However, there are still many unresolved issues in terms of descriptive power and ease of application to new data. Linguistic knowledge such as existing concept dictionaries by experts can be considered a solution to these problems. In this paper, we focus on the Japanese word list by semantic principles constructed by the National Institute for Japanese Language and Linguistics and investigate vectorization by embedding it into hyperbolic space, and the ability and the effectiveness of vectorization of the embedding method for sentiment analysis based on the experimental results.

The purpose of this study is to model the relationship between the intention of remote work and its factors in the provincial areas. In particular, since the popularization of remote work is lower in provincial areas than in metropolitan areas, we focus on partial remote work, which is relatively easy to implement. Therefore, using remote work intention data, we construct not only a model that describes the relationship between the frequency of remote work and those factors, but also a partial remote work model. In the partial remote model, various machine learning models are applied to compare the learning accuracy. As a result, the RF model was able to train almost accurately, and the SVM model was also able to obtain high learning accuracy.

The determination of distance-based toll for urban expressway is discussed to realize the efficient traffic on urban networks in the study. Since the several evaluation indices for social benefit would be obtained from the traffic assignment for urban network, the rather large efforts in complex calculation process should be required. The social benefit evaluation model is created with convolutional neural network approach. As the deep learning may provide the approximation of traffic assignment results, the efficient toll determination problem of urban expressway would be solved practically.

In recent years, various transport services such as sharing services of bicycles and electric kickboards, and on-demand buses have been provided. In addition, the introduction of MaaS, which integrates these transport modes and regards mobility as a service, is also advanced. In the study, an artificial society model is used to examine the appropriate combination of transport services in urban areas. The target area is the southeastern part of Osaka city where services such as Imazato Liner (BRT) and on-demand bus are provided.

Video consumption using smartphones has been increasing along with the growing popularity of video hosting service. On the other hand, due to the vast number of available videos, it is difficult to watch all of them. People are increasingly viewing videos selectively at the peak moments in the videos. Therefore, there is a growing need to extract the peak moments automatically. Existing studies on extracting the peak moments from the video include those focusing on video features using image processing, those focusing on volume and pitch using audio processing, and those analyzing data other than video, such as comments. This study attempts a new approach to extracting the peak moments by focusing on spectrograms representing audio features and applying image processing to them. The evaluation experiments suggest the possibility of automatic extraction of the peak moments in sports videos.

For the automatic creation of math problems, it is necessary to classify not only sentences but also mathematical expressions, but automatic feature extraction methods for sentences containing mathematical expressions have not been established. However, no automatic feature extraction method has been established for sentences containing mathematical expressions. Combining this method with the vectorization of sentences in machine learning, we propose a new method for calculating similarity between math problems.

The FCR method was developed by applying fuzzy logic to solve the problem of centralizing tendency in the rating scale method. In the FCR method, we usually use two items, but there are cases where three or more items are required. Unlike the case of two entities, the analysis method for three things has not been developed yet, except for the fuzzy inference method. In the fuzzy inference method, we measure the degree of conformity of each rule, obtain the output fuzzy set, and perform defuzzy operations on the fuzzy output set to get the singleton value. This study investigates the method for selecting the most dominant rule in fuzzy inference. We also apply our method to analyze how people hear orchestra acoustics differently. In addition, we propose a new defuzzy method called the Mode-Median method.

In this study, a new evaluation method was attempted to investigate onomatopoeia for "so-called" various bipolar pains. The 50 research participants of the elderly group were asked to select onomatopoeia that would represent pain of a particular feature with the bipolar numerical evaluation method. Research participants were then asked to enter a numerical value between 0 and 100 in one of the left and right columns specified by which of the "so-called" bipolar pains they would indicate. Researchers presented participants with 30 of the more than 100 candidates for onomatopoeia that represent specific “so-called” bipolar pain. Then, the ratio of each onomatopoeia selected as the onomatopoeia representing the characteristics of 12 types of pain and the average evaluation value when expressing various types of pain were investigated. The results of this study suggest that there are relatively few Japanese onomatopoeias that represent some of the pain characteristics. The possibility of different numbers of onomatopoeia meeting the specified criteria for different age groups was also investigated. The characteristics and improvements of this new research method were discussed.

In this paper, focusing on the archives of academic journal papers and proceeding papers, we discuss the system development, data migration, system management, and data utilization.

For modeling of preference relations, dominance-based rough set approach (DRSA) was proposed. In the framework of DRSA, classification involving a preference order is considered, and data are expressed by criteria. The inconsistency between the classification and criteria is modeled by approximations of rough set theory. Variable-precision DRSA (VP-DRSA) is an extension of DRSA, in which a certain degree of the inconsistency is tolerated. The lower approximations of VP-DRSA can be interpreted by empirical risk minimization. In this paper, we study formulations of the empirical risk minimization related to VP-DRSA. We propose a new formulation using a nonlinear discriminant function.

In recent years, the demand for data analysis has been increasing in the field of e-sports, which has been developing remarkably. In this study, we aim to find ”conditions and factors closely related to the ranking” from various data related to e-sports by using rough set theory and decision rule analysis. The subject of analysis in this study is PUBG, an online shooting game in battle royale format. As a result, we found common factors that have high contribution to the ranking regardless of the number of team members or the type of map, as well as factors that have different contribution depending on the number of team members and the type of map. In addition, we were able to differentiate our method from other methods of predictive analysis in the analysis of rankings using decision rules.

Variable precision rough set (for short, VPRS) is an extension of Pawlak’s rough set theory to deal with data containing noise or probabilistic information, and various types of reducts are proposed in VPRS. In particular, calculation of reducts preserving lower approximation (called Lβ-reducts) results in calculation of relative reducts of the modified decision tables in Pawlak’s rough set theory. In this paper, we prove that modification of decision tables for computing Lβ-reducts contains redundant procedures.

Abstract: In this paper, we discuss an improved method to quickly calculate relative reducts after appending some objects to the decision table. In general, when a decision table that has already created relative reducts is updated, it is necessary to calculate all relative reducts again. However, recalculation of all relative reducts requires a lot of time and storage area. Therefore, we propose a new method to calculate relative reducts quickly and easily after appending some objects to the decision table.

Recently, artificial-intelligence (AI) coaching systems have become popular because AI automatically evaluates walking motions and suggests training methods. The purpose of this study is to apply such a coaching system to batting form. The system comprises a person-estimation neural network that estimates which person's swing resembles the input swing, a swing-estimation neural network that predicts the estimated person's swing, and a comparison module that compares the estimated and input swings. Herein, we aim to build a system that can compare swings with different speeds.

Vector graphics are composed of meaningful basic components such as lines, curves, and parabolas etc. Therefore, in vector graphics, it is easy to edit each basic component such as resizing, translation, and rotation etc. Because of these advantages, the international e-book file format EPUB recommends saving images in the SVG format. On the other hand, most images are saved electronically in raster format or exist as printed materials. Since an image drawn in raster format is simply a set of pixels, it is not easy to divide the image into its basic components. Therefore, we propose a semantic segmentation method for converting line drawings in raster format into vector format and verify its effectiveness through computer experiments. The model proposed in this paper outperforms a conventional model in terms of both extraction accuracy and computational processing speed.