Pub Date : 2019-03-16DOI: 10.1109/ICTEMSYS.2019.8695969
Sukkpranhachai Gatesichapakorn, M. Ruchanurucks, P. Bunnun, T. Isshiki
This paper presents a pose planning method for ROS-based mobile robot equipped with an onboard computer. The system aims for archiving a remote 3D reconstruction using an onboard RGB-D camera and mobile robots autonomously. To plan a robot pose with a good view point for fixed position of an onboard camera configuration is a task we are addressing in this work. The proposed method is just a part of our system to find a good view point before performing 3D reconstruction tasks. Such system is suitable for a low-power onboard computer in cooperating with a remote server to support for rich computational tasks. A low bandwidth data stream between the onboard computer and the server is used most of time while a high bandwidth data will just be used when needed. Our method uses basic triangulation and transformation to find a good view point based on reference surface points. Reference surface points are extracted by using a cost value from ROS costmap data. The method is implemented and tested in a simulation software and realizing ROS environment. Outcomes with output from camera and visualization software are observed and evaluated.
{"title":"ROS-Based Mobile Robot Pose Planning for a Good View of an Onboard Camera using Costmap","authors":"Sukkpranhachai Gatesichapakorn, M. Ruchanurucks, P. Bunnun, T. Isshiki","doi":"10.1109/ICTEMSYS.2019.8695969","DOIUrl":"https://doi.org/10.1109/ICTEMSYS.2019.8695969","url":null,"abstract":"This paper presents a pose planning method for ROS-based mobile robot equipped with an onboard computer. The system aims for archiving a remote 3D reconstruction using an onboard RGB-D camera and mobile robots autonomously. To plan a robot pose with a good view point for fixed position of an onboard camera configuration is a task we are addressing in this work. The proposed method is just a part of our system to find a good view point before performing 3D reconstruction tasks. Such system is suitable for a low-power onboard computer in cooperating with a remote server to support for rich computational tasks. A low bandwidth data stream between the onboard computer and the server is used most of time while a high bandwidth data will just be used when needed. Our method uses basic triangulation and transformation to find a good view point based on reference surface points. Reference surface points are extracted by using a cost value from ROS costmap data. The method is implemented and tested in a simulation software and realizing ROS environment. Outcomes with output from camera and visualization software are observed and evaluated.","PeriodicalId":220516,"journal":{"name":"2019 10th International Conference of Information and Communication Technology for Embedded Systems (IC-ICTES)","volume":"64 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129542004","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-03-16DOI: 10.1109/ICTEMSYS.2019.8695963
Nitchakun Kantasewi, S. Marukatat, S. Thainimit, Okumura Manabu
Q-learning is a popular reinforcement learning technique for solving shortest path (STP) problem. In a maze with multiple sub-tasks such as collecting treasures and avoiding traps, it has been observed that the Q-learning converges to the optimal path. However, the sum of obtained rewards along the path in average is moderate. This paper proposes Multi-Q-Table Q-learning to address a problem of low average sum of rewards. The proposed method constructs a new Q-table whenever a sub-goal is reached. This modification let an agent to learn that the sub-reward is already collect and it can be obtained only once. Our experimental results show that a modified algorithm can achieve an optimal answer to collect all treasures (positive rewards), avoid pit and reach goal with the shortest path. With a small size of maze, the proposed algorithm uses the larger amount of time to achieved optimal solution compared to the conventional Q-learning.
{"title":"Multi Q-Table Q-Learning","authors":"Nitchakun Kantasewi, S. Marukatat, S. Thainimit, Okumura Manabu","doi":"10.1109/ICTEMSYS.2019.8695963","DOIUrl":"https://doi.org/10.1109/ICTEMSYS.2019.8695963","url":null,"abstract":"Q-learning is a popular reinforcement learning technique for solving shortest path (STP) problem. In a maze with multiple sub-tasks such as collecting treasures and avoiding traps, it has been observed that the Q-learning converges to the optimal path. However, the sum of obtained rewards along the path in average is moderate. This paper proposes Multi-Q-Table Q-learning to address a problem of low average sum of rewards. The proposed method constructs a new Q-table whenever a sub-goal is reached. This modification let an agent to learn that the sub-reward is already collect and it can be obtained only once. Our experimental results show that a modified algorithm can achieve an optimal answer to collect all treasures (positive rewards), avoid pit and reach goal with the shortest path. With a small size of maze, the proposed algorithm uses the larger amount of time to achieved optimal solution compared to the conventional Q-learning.","PeriodicalId":220516,"journal":{"name":"2019 10th International Conference of Information and Communication Technology for Embedded Systems (IC-ICTES)","volume":"43 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124780855","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-03-16DOI: 10.1109/ICTEMSYS.2019.8695966
Pruttapon Maolanon, K. Sukvichai, N. Chayopitak, Atsushi Takahashi
In order to make autonomous home service robot able to navigate through its environment, one requires a surrounding map and the robot’s location. The Simultaneous Localization And Mapping or SLAM is the method that gathers information from an interested unknown environment, and creates a map and also predicts robot position at the same time. SLAM map is not enough for robot builder companies to sell their service robot because the robots cannot recognize the room in house without complex setup from the experts. The robot cannot be opened from its package and immediately ready to be used. In this research, one method to overcome this issue is proposed by enhancing SLAM algorithm with furniture and household object detection CNN network in order to increase robot ability. Robot will create maps by using a Laser Scan Matcher based on visual SLAM using 3D Orbbec camera. YOLO v3 tiny network is selected as the CNN detector for localize and classify household objects and furnitures in a house. Furnitures and objects images are used to train the CNN networks separately in desktop PC and are installed into the robot after training is finished. CNN detector is combined with SLAM algorithm via ROS. Now, SLAM map can be generated and room can be detected simultaneously in the unknown environment automatically. Finally, experiment is conducted to test the proposed method.
{"title":"Indoor Room Identify and Mapping with Virtual based SLAM using Furnitures and Household Objects Relationship based on CNNs","authors":"Pruttapon Maolanon, K. Sukvichai, N. Chayopitak, Atsushi Takahashi","doi":"10.1109/ICTEMSYS.2019.8695966","DOIUrl":"https://doi.org/10.1109/ICTEMSYS.2019.8695966","url":null,"abstract":"In order to make autonomous home service robot able to navigate through its environment, one requires a surrounding map and the robot’s location. The Simultaneous Localization And Mapping or SLAM is the method that gathers information from an interested unknown environment, and creates a map and also predicts robot position at the same time. SLAM map is not enough for robot builder companies to sell their service robot because the robots cannot recognize the room in house without complex setup from the experts. The robot cannot be opened from its package and immediately ready to be used. In this research, one method to overcome this issue is proposed by enhancing SLAM algorithm with furniture and household object detection CNN network in order to increase robot ability. Robot will create maps by using a Laser Scan Matcher based on visual SLAM using 3D Orbbec camera. YOLO v3 tiny network is selected as the CNN detector for localize and classify household objects and furnitures in a house. Furnitures and objects images are used to train the CNN networks separately in desktop PC and are installed into the robot after training is finished. CNN detector is combined with SLAM algorithm via ROS. Now, SLAM map can be generated and room can be detected simultaneously in the unknown environment automatically. Finally, experiment is conducted to test the proposed method.","PeriodicalId":220516,"journal":{"name":"2019 10th International Conference of Information and Communication Technology for Embedded Systems (IC-ICTES)","volume":"28 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122347169","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-03-16DOI: 10.1109/ICTEMSYS.2019.8695964
P. Charuchinda, T. Kasetkasem, I. Kumazawa, T. Chanwimaluang
Traditionally, the land cover mapping process needs a ground data to be collected with high precision in both class labeling and spatial locations. To collect enough, high precise ground data require resources. As a result, we proposed an approach for building an image classification based on the class activation map (CAM) where the goal is not to identify the relationship between each pixel and a class label, but to identify whether each sub-images contain the class of interest or not. The output of the class activation map is the filter responds where pixels with high respond are likely to belong to the class of interest. We examined the performance on a LAND-SAT 8 and found. The result of CAM showed that the proposed method achieves high accuracy in identifying whether a sub-image contains the class of interest or not. However, the precision in localizing the class is relatively moderate.
{"title":"On Building Detection Using the Class Activation Map: Case Study on a Landsat8 Image","authors":"P. Charuchinda, T. Kasetkasem, I. Kumazawa, T. Chanwimaluang","doi":"10.1109/ICTEMSYS.2019.8695964","DOIUrl":"https://doi.org/10.1109/ICTEMSYS.2019.8695964","url":null,"abstract":"Traditionally, the land cover mapping process needs a ground data to be collected with high precision in both class labeling and spatial locations. To collect enough, high precise ground data require resources. As a result, we proposed an approach for building an image classification based on the class activation map (CAM) where the goal is not to identify the relationship between each pixel and a class label, but to identify whether each sub-images contain the class of interest or not. The output of the class activation map is the filter responds where pixels with high respond are likely to belong to the class of interest. We examined the performance on a LAND-SAT 8 and found. The result of CAM showed that the proposed method achieves high accuracy in identifying whether a sub-image contains the class of interest or not. However, the precision in localizing the class is relatively moderate.","PeriodicalId":220516,"journal":{"name":"2019 10th International Conference of Information and Communication Technology for Embedded Systems (IC-ICTES)","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128424899","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-03-01DOI: 10.1109/ICTEMSYS.2019.8695951
Chatchawal Sangkeettrakarn, C. Haruechaiyasak, T. Theeramunkong
Machine understanding research aims to build machine intelligences. To make a machine understand, precise concepts are necessary. Numerous domains contain vague meanings when making decisions, such as a diagnosis or a legal interpretation. Once an artificial intelligence pretends to be human while dealing with imprecise data, a fuzziness in knowledges must be detected before constructing.This paper presents the methodology to detect a fuzziness in Thai law texts using a deep learning method. The experiments are designed to compare the performances of four well-known text classification methods, namely Decision Tree, Random Forest, Support Vector Machine, and Convolutional Neural Network. The fuzziness in this study refers to an imprecise meaning in law texts which may be ambiguous when interpreted by a machine. We built a labelled corpus from four Thai Law codes namely 1) The Criminal Code 2) The Criminal Procedure Code 3) The Civil and Commercial Code and 4) The Civil Procedure Code. We proposed three conditions to identify the fuzziness, i.e. 1) a decision depends on a judge’s opinion 2) a decision that requires the production of evidence and 3) a decision which refers to other sections. The results of the experiment show that a Convolutional Neural Network significantly outperforms the others with 97.54% accuracy in comparison of all the dataset.
{"title":"Fuzziness Detection in Thai Law Texts Using Deep Learning","authors":"Chatchawal Sangkeettrakarn, C. Haruechaiyasak, T. Theeramunkong","doi":"10.1109/ICTEMSYS.2019.8695951","DOIUrl":"https://doi.org/10.1109/ICTEMSYS.2019.8695951","url":null,"abstract":"Machine understanding research aims to build machine intelligences. To make a machine understand, precise concepts are necessary. Numerous domains contain vague meanings when making decisions, such as a diagnosis or a legal interpretation. Once an artificial intelligence pretends to be human while dealing with imprecise data, a fuzziness in knowledges must be detected before constructing.This paper presents the methodology to detect a fuzziness in Thai law texts using a deep learning method. The experiments are designed to compare the performances of four well-known text classification methods, namely Decision Tree, Random Forest, Support Vector Machine, and Convolutional Neural Network. The fuzziness in this study refers to an imprecise meaning in law texts which may be ambiguous when interpreted by a machine. We built a labelled corpus from four Thai Law codes namely 1) The Criminal Code 2) The Criminal Procedure Code 3) The Civil and Commercial Code and 4) The Civil Procedure Code. We proposed three conditions to identify the fuzziness, i.e. 1) a decision depends on a judge’s opinion 2) a decision that requires the production of evidence and 3) a decision which refers to other sections. The results of the experiment show that a Convolutional Neural Network significantly outperforms the others with 97.54% accuracy in comparison of all the dataset.","PeriodicalId":220516,"journal":{"name":"2019 10th International Conference of Information and Communication Technology for Embedded Systems (IC-ICTES)","volume":"88 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122705847","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-03-01DOI: 10.1109/ictemsys.2019.8695922
{"title":"IC-ICTES 2019 Committees","authors":"","doi":"10.1109/ictemsys.2019.8695922","DOIUrl":"https://doi.org/10.1109/ictemsys.2019.8695922","url":null,"abstract":"","PeriodicalId":220516,"journal":{"name":"2019 10th International Conference of Information and Communication Technology for Embedded Systems (IC-ICTES)","volume":"16 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125668984","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-03-01DOI: 10.1109/ictemsys.2019.8695910
{"title":"IC-ICTES 2019 Reviewer List","authors":"","doi":"10.1109/ictemsys.2019.8695910","DOIUrl":"https://doi.org/10.1109/ictemsys.2019.8695910","url":null,"abstract":"","PeriodicalId":220516,"journal":{"name":"2019 10th International Conference of Information and Communication Technology for Embedded Systems (IC-ICTES)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130805057","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-03-01DOI: 10.1109/ICTEMSYS.2019.8695960
S. Sakphrom, S. Korkua
In order to support and provide more accurate information which is needed to better operate and manage in hydrological applications, this paper proposes the simplified estimation of the stream discharge based on width, depth and water flowing rate measurements. In this paper, the NB-IoT deployment composed of width sensor, depth sensor and water flow sensor is designed and implemented for water resource management such as stream, canal, river, etc. The proposed hydrological WSN comsists of two main parts: 1) a remote control via mobile application and 2) a small boat built-in all sensing devices. As the focus of this paper, the hardware selections are discussed in detail. Based on the Arduino UNO microcontroller platform, the system can collect data and send out all measured data through Bluetooth communication. Moreover, it is controlled by features of the localization and then can represent its positioning with the Google mapping application via mobile device. Experimental results of the simplified stream discharge estimation verified that the proposed NB-IoT system can accurately monitor the width and depth parameters and also estimate the stream discharge properly.
{"title":"Simplified Stream Discharge Estimation for Hydrological Application based on NB-IoT Deployment","authors":"S. Sakphrom, S. Korkua","doi":"10.1109/ICTEMSYS.2019.8695960","DOIUrl":"https://doi.org/10.1109/ICTEMSYS.2019.8695960","url":null,"abstract":"In order to support and provide more accurate information which is needed to better operate and manage in hydrological applications, this paper proposes the simplified estimation of the stream discharge based on width, depth and water flowing rate measurements. In this paper, the NB-IoT deployment composed of width sensor, depth sensor and water flow sensor is designed and implemented for water resource management such as stream, canal, river, etc. The proposed hydrological WSN comsists of two main parts: 1) a remote control via mobile application and 2) a small boat built-in all sensing devices. As the focus of this paper, the hardware selections are discussed in detail. Based on the Arduino UNO microcontroller platform, the system can collect data and send out all measured data through Bluetooth communication. Moreover, it is controlled by features of the localization and then can represent its positioning with the Google mapping application via mobile device. Experimental results of the simplified stream discharge estimation verified that the proposed NB-IoT system can accurately monitor the width and depth parameters and also estimate the stream discharge properly.","PeriodicalId":220516,"journal":{"name":"2019 10th International Conference of Information and Communication Technology for Embedded Systems (IC-ICTES)","volume":"23 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122067646","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-03-01DOI: 10.1109/ICTEMSYS.2019.8695968
Kanitkorn Khanchuea, Rawat Siripokarpirom
This paper presents the design and implementation of a multi-protocol gateway that relies on a multi-hop wireless network for smart home and building automation applications. This paper describes how to construct such a multi-hop tree-based wireless network to coexist with ZigBee mesh networks, using low-cost commodity 2.4GHz WiFi/BLE SoC modules. The gateway supports both wired and wireless connectivities to other sensor nodes, including the RS485/Modbus fieldbus and wireless networks based on two different protocols, namely the ESP-Now peer-to-peer wireless protocol developed by Espressif Systems and the ZigBee protocol. In this work, the ESP-NOW protocol was utilized to construct the core low-power multi-hop wireless network, whereas the ZigBee standard was used to build subnetworks of sensor nodes. A single-board computer (SBC), running an embedded Linux operating system, was chosen as a platform for implementing the proposed IoT gateway which utilized the MQTT protocol for message delivery. Field experiments were conducted to evaluate the performance of the ESP-Now multi-hop wireless network, comprised of up to 5 hops. To illustrate the functionality of the proposed gateway, a use case was presented, in which a building automation system prototype was developed for control and management of air-conditioners and AC power meter units.
{"title":"A Multi-Protocol IoT Gateway and WiFi/BLE Sensor Nodes for Smart Home and Building Automation: Design and Implementation","authors":"Kanitkorn Khanchuea, Rawat Siripokarpirom","doi":"10.1109/ICTEMSYS.2019.8695968","DOIUrl":"https://doi.org/10.1109/ICTEMSYS.2019.8695968","url":null,"abstract":"This paper presents the design and implementation of a multi-protocol gateway that relies on a multi-hop wireless network for smart home and building automation applications. This paper describes how to construct such a multi-hop tree-based wireless network to coexist with ZigBee mesh networks, using low-cost commodity 2.4GHz WiFi/BLE SoC modules. The gateway supports both wired and wireless connectivities to other sensor nodes, including the RS485/Modbus fieldbus and wireless networks based on two different protocols, namely the ESP-Now peer-to-peer wireless protocol developed by Espressif Systems and the ZigBee protocol. In this work, the ESP-NOW protocol was utilized to construct the core low-power multi-hop wireless network, whereas the ZigBee standard was used to build subnetworks of sensor nodes. A single-board computer (SBC), running an embedded Linux operating system, was chosen as a platform for implementing the proposed IoT gateway which utilized the MQTT protocol for message delivery. Field experiments were conducted to evaluate the performance of the ESP-Now multi-hop wireless network, comprised of up to 5 hops. To illustrate the functionality of the proposed gateway, a use case was presented, in which a building automation system prototype was developed for control and management of air-conditioners and AC power meter units.","PeriodicalId":220516,"journal":{"name":"2019 10th International Conference of Information and Communication Technology for Embedded Systems (IC-ICTES)","volume":"108 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133208126","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-03-01DOI: 10.1109/ICTEMSYS.2019.8695967
K. Sukvichai, Pruttapon Maolanon, Kittinon sawanyawat, Warayut Muknumporn
Foods are important things to human lives, especially for elderly or diabetics. Tradition nutrition book is not the effective way for people to use and not cover all kind of foods. Most of the food nutrition in the book focused on Western dishes not Asian dishes. This research proposed the new way to categorized Thai fast food dishes, classified and localized the ingredients in each dish. Convolutional Neural Networks (CNNs) are used to achieve these tasks. MobileNet is used as food categorizer while You Only Look Once (YOLO) network works as the ingredients classifier and localizer. Then, ingredients in the pictures are cropped and passed through traditional image processing to calculate area and compared with real ingredient’s dimension. Non-uniform shape ingredients are segmented, then, the nutrition of the dish can be calculated. Finally, the networks are transferred in to Raspberry Pi 3 platform to simulate limited resources and calculation power platform likes in a mobile phone. The networks in Raspberry Pi 3 produce good prediction accuracy but slow speed. PeachPy is introduced to speed up the network and it can run at 3.3 seconds per food image.
食物对人类的生活很重要，尤其是对老年人和糖尿病患者。传统的营养书籍并不是人们使用的有效方式，也没有涵盖所有的食物。书中大部分的食物营养都集中在西餐上，而不是亚洲菜。本研究提出了泰式快餐菜肴分类的新方法，对每道菜中的食材进行分类和本土化。卷积神经网络(cnn)被用来完成这些任务。MobileNet用作食品分类器，而You Only Look Once (YOLO)网络用作成分分类器和本地化器。然后，对图片中的成分进行裁剪，通过传统的图像处理计算面积，并与真实成分的尺寸进行比较。将形状不均匀的食材进行分割，计算出菜肴的营养成分。最后，将网络传输到树莓派3平台上，模拟类似于手机的有限资源和计算能力平台。树莓派3中的网络具有良好的预测精度，但速度较慢。引入PeachPy是为了加快网络速度，它可以以3.3秒的速度运行每个食物图像。
{"title":"Food categories classification and Ingredients estimation using CNNs on Raspberry Pi 3","authors":"K. Sukvichai, Pruttapon Maolanon, Kittinon sawanyawat, Warayut Muknumporn","doi":"10.1109/ICTEMSYS.2019.8695967","DOIUrl":"https://doi.org/10.1109/ICTEMSYS.2019.8695967","url":null,"abstract":"Foods are important things to human lives, especially for elderly or diabetics. Tradition nutrition book is not the effective way for people to use and not cover all kind of foods. Most of the food nutrition in the book focused on Western dishes not Asian dishes. This research proposed the new way to categorized Thai fast food dishes, classified and localized the ingredients in each dish. Convolutional Neural Networks (CNNs) are used to achieve these tasks. MobileNet is used as food categorizer while You Only Look Once (YOLO) network works as the ingredients classifier and localizer. Then, ingredients in the pictures are cropped and passed through traditional image processing to calculate area and compared with real ingredient’s dimension. Non-uniform shape ingredients are segmented, then, the nutrition of the dish can be calculated. Finally, the networks are transferred in to Raspberry Pi 3 platform to simulate limited resources and calculation power platform likes in a mobile phone. The networks in Raspberry Pi 3 produce good prediction accuracy but slow speed. PeachPy is introduced to speed up the network and it can run at 3.3 seconds per food image.","PeriodicalId":220516,"journal":{"name":"2019 10th International Conference of Information and Communication Technology for Embedded Systems (IC-ICTES)","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114381278","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：