Pub Date : 2018-11-01DOI: 10.1109/IPTA.2018.8608154
Sara Akodad, L. Bombrun, C. Yaacoub, Y. Berthoumieu, C. Germain
This paper introduces an image classification method based on the encoding of a set of covariance matrices. This encoding relies on Fisher vectors adapted to the log-Euclidean metric: the log-Euclidean Fisher vectors (LE FV). This approach is next extended to full local Gaussian descriptors composed by a set of local mean vectors and local covariance matrices. For that, the local Gaussian model is transformed to a zero-mean Gaussian model with an augmented covariance matrix. All these approaches are used to encode handcrafted or deep learning features. Finally, they are applied in a remote sensing application on the UC Merced dataset which consists in classifying land cover images. A sensitivity analysis is carried out to evaluate the potential of the proposed LE FV.
{"title":"Image classification based on log-Euclidean Fisher Vectors for covariance matrix descriptors","authors":"Sara Akodad, L. Bombrun, C. Yaacoub, Y. Berthoumieu, C. Germain","doi":"10.1109/IPTA.2018.8608154","DOIUrl":"https://doi.org/10.1109/IPTA.2018.8608154","url":null,"abstract":"This paper introduces an image classification method based on the encoding of a set of covariance matrices. This encoding relies on Fisher vectors adapted to the log-Euclidean metric: the log-Euclidean Fisher vectors (LE FV). This approach is next extended to full local Gaussian descriptors composed by a set of local mean vectors and local covariance matrices. For that, the local Gaussian model is transformed to a zero-mean Gaussian model with an augmented covariance matrix. All these approaches are used to encode handcrafted or deep learning features. Finally, they are applied in a remote sensing application on the UC Merced dataset which consists in classifying land cover images. A sensitivity analysis is carried out to evaluate the potential of the proposed LE FV.","PeriodicalId":272294,"journal":{"name":"2018 Eighth International Conference on Image Processing Theory, Tools and Applications (IPTA)","volume":"20 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129340370","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 : 2018-11-01DOI: 10.1109/IPTA.2018.8608161
H. Ju, Zhigang Liu, Yang Wang
Most anomaly detection methods for hyperspectral image (HSI) have focused on the spectral information while ignoring the spatial information. In this paper, a novel anomaly detection method has been proposed in which the spatial information has been incorporated. Firstly, the dual windows are established to estimate the background of pixel under test (PUT). Secondly, the spectral distance is calculated between PUT and its background to measure its spectral anomaly degree. Then, the principal component analysis is performed on HSI and the spatial anomaly degree of PUT is measured on the first component by comparing the spatial structure similarity between PUT and its background. Lastly, combining the spectral anomaly degree and the spatial anomaly degree, the anomaly degree of PUT is obtained. Experimental results on two hyperspectral datasets confirm the proposed method is superior to three commonly used state-of-the-art anomaly detection methods in suppressing the background and detecting anomalies and is also quite robust to noise.
{"title":"Hyperspetral Anomaly Detection Incorporating Spatial Information","authors":"H. Ju, Zhigang Liu, Yang Wang","doi":"10.1109/IPTA.2018.8608161","DOIUrl":"https://doi.org/10.1109/IPTA.2018.8608161","url":null,"abstract":"Most anomaly detection methods for hyperspectral image (HSI) have focused on the spectral information while ignoring the spatial information. In this paper, a novel anomaly detection method has been proposed in which the spatial information has been incorporated. Firstly, the dual windows are established to estimate the background of pixel under test (PUT). Secondly, the spectral distance is calculated between PUT and its background to measure its spectral anomaly degree. Then, the principal component analysis is performed on HSI and the spatial anomaly degree of PUT is measured on the first component by comparing the spatial structure similarity between PUT and its background. Lastly, combining the spectral anomaly degree and the spatial anomaly degree, the anomaly degree of PUT is obtained. Experimental results on two hyperspectral datasets confirm the proposed method is superior to three commonly used state-of-the-art anomaly detection methods in suppressing the background and detecting anomalies and is also quite robust to noise.","PeriodicalId":272294,"journal":{"name":"2018 Eighth International Conference on Image Processing Theory, Tools and Applications (IPTA)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129667426","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 : 2018-11-01DOI: 10.1109/IPTA.2018.8608140
Zhan Qi, Maojun Li, Chongpei Liu, M. Zhao, Manyi Long
We propose a real-time detection method for vehicle queue length detection at intersections based on image processing technology. Firstly, we acquire the image of vehicle queue at the intersection, and apply the automatic brightness adjustment algorithm and lane line detection algorithm to reduce the affect of different light intensity and camera shake on the image respectively. And then, the preprocessed image is subtracted from the background image to obtain the foreground image of queue vehicle. Finally, we detect the vehicle queue length by the middle line, and the actual vehicle queue length is measured by the camera calibration method. The experimental results show that the proposed method has high accuracy rate and is fast enough for practical application.
{"title":"A Measurement Method for Vehicle Queue Length of Intersection Based on Image Processing","authors":"Zhan Qi, Maojun Li, Chongpei Liu, M. Zhao, Manyi Long","doi":"10.1109/IPTA.2018.8608140","DOIUrl":"https://doi.org/10.1109/IPTA.2018.8608140","url":null,"abstract":"We propose a real-time detection method for vehicle queue length detection at intersections based on image processing technology. Firstly, we acquire the image of vehicle queue at the intersection, and apply the automatic brightness adjustment algorithm and lane line detection algorithm to reduce the affect of different light intensity and camera shake on the image respectively. And then, the preprocessed image is subtracted from the background image to obtain the foreground image of queue vehicle. Finally, we detect the vehicle queue length by the middle line, and the actual vehicle queue length is measured by the camera calibration method. The experimental results show that the proposed method has high accuracy rate and is fast enough for practical application.","PeriodicalId":272294,"journal":{"name":"2018 Eighth International Conference on Image Processing Theory, Tools and Applications (IPTA)","volume":"28 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129134462","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 : 2018-11-01DOI: 10.1109/IPTA.2018.8608155
Xiaoyue Jiang, Zhongyun Hu, Yue Ni
Shadows bring great challenges but also play essential roles in image understanding. Most recent shadow detection methods are based on patches, then further reasoning method is required for the obtaining of a completed shadow detection result for an image. In this paper, an injection network is proposed to detect shadow regions for the whole image directly. In order to maintain as many as details, the skip structure is applied to directly inject the details from convolutional layers to de-convolutional layers. Meanwhile, a weighted loss function is proposed for the network training. With this adapted loss function, the network becomes more sensitive to errors of shadow regions. Thus the proposed network can focus on the learning of robust shadow features. Furthermore, a shadow refinement method is proposed to optimize the boundary region of shadows. In the experiments, the proposed methods are extensively evaluated on two popular datasets and shown better performance on shadow detection compared with current methods.
{"title":"InNet: Learning to Detect Shadows with Injection Network","authors":"Xiaoyue Jiang, Zhongyun Hu, Yue Ni","doi":"10.1109/IPTA.2018.8608155","DOIUrl":"https://doi.org/10.1109/IPTA.2018.8608155","url":null,"abstract":"Shadows bring great challenges but also play essential roles in image understanding. Most recent shadow detection methods are based on patches, then further reasoning method is required for the obtaining of a completed shadow detection result for an image. In this paper, an injection network is proposed to detect shadow regions for the whole image directly. In order to maintain as many as details, the skip structure is applied to directly inject the details from convolutional layers to de-convolutional layers. Meanwhile, a weighted loss function is proposed for the network training. With this adapted loss function, the network becomes more sensitive to errors of shadow regions. Thus the proposed network can focus on the learning of robust shadow features. Furthermore, a shadow refinement method is proposed to optimize the boundary region of shadows. In the experiments, the proposed methods are extensively evaluated on two popular datasets and shown better performance on shadow detection compared with current methods.","PeriodicalId":272294,"journal":{"name":"2018 Eighth International Conference on Image Processing Theory, Tools and Applications (IPTA)","volume":"22 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114247552","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}
In this paper, we present a new, large-scale database on underwater image, which is called the NWPU underwater image database. This database contains 6240 underwater images of 40 objects. Each object is captured with 6 different levels of turbidity water, 4 lighting conditions and 6 different distances. Among them, we use the underwater images with turbidity value of 0 as Ground-truth. In addition, we captured the shadowless image of the object in the air and clear water. Different from other underwater databases, we capture underwater images with real high turbidity lake water instead of simulating the turbidity of water. This method ensures that the underwater images we captured are as close as possible to the real environment. We have given the database baseline which contains multi-scale Retinex with color restore (MSRCR) algorithms for enhancing images and four commonly used image quality evaluation criteria, including two full-references and two no-references methods. The four image quality evaluation methods include two no-reference and two full reference.
{"title":"A New Database for Evaluating Underwater Image Processing Methods","authors":"Yupeng Ma, Xiaoyi Feng, Lujing Chao, Dong Huang, Zhaoqiang Xia, Xiaoyue Jiang","doi":"10.1109/IPTA.2018.8608131","DOIUrl":"https://doi.org/10.1109/IPTA.2018.8608131","url":null,"abstract":"In this paper, we present a new, large-scale database on underwater image, which is called the NWPU underwater image database. This database contains 6240 underwater images of 40 objects. Each object is captured with 6 different levels of turbidity water, 4 lighting conditions and 6 different distances. Among them, we use the underwater images with turbidity value of 0 as Ground-truth. In addition, we captured the shadowless image of the object in the air and clear water. Different from other underwater databases, we capture underwater images with real high turbidity lake water instead of simulating the turbidity of water. This method ensures that the underwater images we captured are as close as possible to the real environment. We have given the database baseline which contains multi-scale Retinex with color restore (MSRCR) algorithms for enhancing images and four commonly used image quality evaluation criteria, including two full-references and two no-references methods. The four image quality evaluation methods include two no-reference and two full reference.","PeriodicalId":272294,"journal":{"name":"2018 Eighth International Conference on Image Processing Theory, Tools and Applications (IPTA)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132745118","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 : 2018-11-01DOI: 10.1109/IPTA.2018.8608162
Tian Lan, Xiaoyi Feng, Lei Li, Zhaoqiang Xia
Trademarks are intellectual and industrial properties developed under the commodity economy, representing reputation, quality and reliability of firms. Therefore, in order to prevent the registration of new trademarks from having a high-degree similarity with registered ones, we propose a new trademark retrieval method. Based on the fact that the shape and color of a trademark are varied, our proposed method combines a metric convolutional neural network (CNN) and conventional hand-crafted features to describe the trademark images. More specifically, we first train the CNN based on Siamese and Triplet structures, and then extract the hand-crafted features from convolutional feature maps. For this research, we utilize a challenging trademark dataset that contains 7139 various color or gray images. Besides, extensive experiments on our dataset and the METU public dataset demonstrate the effectiveness of our method in trademark retrieval and achieve the state-of-the-art performance compared to traditional countermeasures.
{"title":"Similar Trademark Image Retrieval Based on Convolutional Neural Network and Constraint Theory","authors":"Tian Lan, Xiaoyi Feng, Lei Li, Zhaoqiang Xia","doi":"10.1109/IPTA.2018.8608162","DOIUrl":"https://doi.org/10.1109/IPTA.2018.8608162","url":null,"abstract":"Trademarks are intellectual and industrial properties developed under the commodity economy, representing reputation, quality and reliability of firms. Therefore, in order to prevent the registration of new trademarks from having a high-degree similarity with registered ones, we propose a new trademark retrieval method. Based on the fact that the shape and color of a trademark are varied, our proposed method combines a metric convolutional neural network (CNN) and conventional hand-crafted features to describe the trademark images. More specifically, we first train the CNN based on Siamese and Triplet structures, and then extract the hand-crafted features from convolutional feature maps. For this research, we utilize a challenging trademark dataset that contains 7139 various color or gray images. Besides, extensive experiments on our dataset and the METU public dataset demonstrate the effectiveness of our method in trademark retrieval and achieve the state-of-the-art performance compared to traditional countermeasures.","PeriodicalId":272294,"journal":{"name":"2018 Eighth International Conference on Image Processing Theory, Tools and Applications (IPTA)","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115834218","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}
Pedestrian detection, which has broad application prospects in video security, robotics and self-driving vehicles etc., is one of the most important research fields in computer vision. Recently, deep learning methods, e.g., Region Proposal Network (RPN), have achieved major performance improvements in pedestrian detection. In order to further utilize the deep pedestrian features of RPN, this paper proposes a novel regional proposal network model based on feature fusion (RPN_FeaFus) for pedestrian detection. RPN_FeaFus adopts an asymmetric dual-path deep model, constructed by VGGNet and ZFNet, to extract pedestrian features in different levels, which are further combined through PCA dimension reduction and feature stacking to provide more discriminant representation. Then, the low-dimensional fusion features are adopted to detect the region proposals and train the classifier. Experimental results on three widely used pedestrian detection databases, i.e, Caltech database, Daimler database and TUD database, illuminate that RPN_FeaFus gains obvious performance improvements over its baseline RPN_BF, which is also competitive with the state-of-the-art methods.
{"title":"Pedestrian Detection Using Regional Proposal Network with Feature Fusion","authors":"Xiaogang Lv, Xiaotao Zhang, Yinghua Jiang, Jianxin Zhang","doi":"10.1109/IPTA.2018.8608159","DOIUrl":"https://doi.org/10.1109/IPTA.2018.8608159","url":null,"abstract":"Pedestrian detection, which has broad application prospects in video security, robotics and self-driving vehicles etc., is one of the most important research fields in computer vision. Recently, deep learning methods, e.g., Region Proposal Network (RPN), have achieved major performance improvements in pedestrian detection. In order to further utilize the deep pedestrian features of RPN, this paper proposes a novel regional proposal network model based on feature fusion (RPN_FeaFus) for pedestrian detection. RPN_FeaFus adopts an asymmetric dual-path deep model, constructed by VGGNet and ZFNet, to extract pedestrian features in different levels, which are further combined through PCA dimension reduction and feature stacking to provide more discriminant representation. Then, the low-dimensional fusion features are adopted to detect the region proposals and train the classifier. Experimental results on three widely used pedestrian detection databases, i.e, Caltech database, Daimler database and TUD database, illuminate that RPN_FeaFus gains obvious performance improvements over its baseline RPN_BF, which is also competitive with the state-of-the-art methods.","PeriodicalId":272294,"journal":{"name":"2018 Eighth International Conference on Image Processing Theory, Tools and Applications (IPTA)","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124491746","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}
The automatic recognition of spontaneous facial micro-expressions becomes prevalent as it reveals the actual emotion of humans. However, handcrafted features employed for recognizing micro-expressions are designed for general applications and thus cannot well capture the subtle facial deformations of micro-expressions. To address this problem, we propose an end-to-end deep learning framework to suit the particular needs of micro-expression recognition (MER). In the deep model, re- current convolutional networks are utilized to learn the representation of subtle changes from image sequences. To guarantee the learning of deep model, we present a temporal jittering procedure to greatly enrich the training samples. Through performing the experiments on three spontaneous micro-expression datasets, i.e., SMIC, CASME, and CASME2, we verify the effectiveness of our proposed MER approach.
{"title":"Spontaneous Facial Micro-expression Recognition via Deep Convolutional Network","authors":"Zhaoqiang Xia, Xiaoyi Feng, Xiaopeng Hong, Guoying Zhao","doi":"10.1109/IPTA.2018.8608119","DOIUrl":"https://doi.org/10.1109/IPTA.2018.8608119","url":null,"abstract":"The automatic recognition of spontaneous facial micro-expressions becomes prevalent as it reveals the actual emotion of humans. However, handcrafted features employed for recognizing micro-expressions are designed for general applications and thus cannot well capture the subtle facial deformations of micro-expressions. To address this problem, we propose an end-to-end deep learning framework to suit the particular needs of micro-expression recognition (MER). In the deep model, re- current convolutional networks are utilized to learn the representation of subtle changes from image sequences. To guarantee the learning of deep model, we present a temporal jittering procedure to greatly enrich the training samples. Through performing the experiments on three spontaneous micro-expression datasets, i.e., SMIC, CASME, and CASME2, we verify the effectiveness of our proposed MER approach.","PeriodicalId":272294,"journal":{"name":"2018 Eighth International Conference on Image Processing Theory, Tools and Applications (IPTA)","volume":"131 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121424422","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 : 2018-11-01DOI: 10.1109/IPTA.2018.8608144
Hui Zhang, Guoliang Shi, Li Liu, M. Zhao, Zhicong Liang
The widespread use of barcode technology has led to the complexity of the application scenario. In the traditional barcode recognition method, there is no universal solution to the problems of uneven illumination, distortion, and sheltered. In this paper, the deep learning theory is used to solve the problem of barcode detection under the above situation. And on this basis, the problem of correcting linear distortion Data Matrix code is solved, and the key technology of barcode recognition under complex situation is broken through. After testing, the recognition speed reached 125ms, and the recognition accuracy reached about 93%. The system uses CCD camera to collect pictures, adopts the HALCON to build the processing algorithm, and uses Visual Studio platform to build the software, which realizes the Date Matrix code, Drug Electronic Supervision Code and Product bar code fast and accurate identification on pharmaceutical packaging. The developed system can also detect the rotation angle of Barcode and Data Matrix code, which is favorable for reading the barcode information. The whole process is real-time.
{"title":"Detection and identification method of medical label barcode based on deep learning","authors":"Hui Zhang, Guoliang Shi, Li Liu, M. Zhao, Zhicong Liang","doi":"10.1109/IPTA.2018.8608144","DOIUrl":"https://doi.org/10.1109/IPTA.2018.8608144","url":null,"abstract":"The widespread use of barcode technology has led to the complexity of the application scenario. In the traditional barcode recognition method, there is no universal solution to the problems of uneven illumination, distortion, and sheltered. In this paper, the deep learning theory is used to solve the problem of barcode detection under the above situation. And on this basis, the problem of correcting linear distortion Data Matrix code is solved, and the key technology of barcode recognition under complex situation is broken through. After testing, the recognition speed reached 125ms, and the recognition accuracy reached about 93%. The system uses CCD camera to collect pictures, adopts the HALCON to build the processing algorithm, and uses Visual Studio platform to build the software, which realizes the Date Matrix code, Drug Electronic Supervision Code and Product bar code fast and accurate identification on pharmaceutical packaging. The developed system can also detect the rotation angle of Barcode and Data Matrix code, which is favorable for reading the barcode information. The whole process is real-time.","PeriodicalId":272294,"journal":{"name":"2018 Eighth International Conference on Image Processing Theory, Tools and Applications (IPTA)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122353010","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 : 2018-11-01DOI: 10.1109/IPTA.2018.8608158
M. Zhao, Hui Zhang, Li Liu, Zhicong Liang, Guang Deng
At present, the detection of pharmaceutical injection products is a quite important step in the pharmaceutical manufacturing, as it has the direct related to the quality of medical product quality. Aiming at the difficulty that liquid particle has a smaller pixel point in the high resolution image of detection of pharmaceutical liquid particle, hence consider combined with deep neural network and clustering algorithm for detection and localization of little particle, and a processing method combining single frame images with multi-frame images was proposed to identifying liquid particle. Firstly, the single-frame image is detected by using Faster-RCNN deep neural network, and it can obtain the detection result of the 8-frame sequence image. Then hierarchical clustering and K-means clustering algorithm are used for clustering to obtain the same target motion area. In this way, liquid particle can be more accurately identified and the accuracy of detection can be greatly improved. The experimental results show that the accuracy of detection and recognition of foreign substances in liquid medicine is improved by more than 10% on average.
{"title":"Joint Deep Learning and Clustering Algorithm for Liquid Particle Detection of Pharmaceutical Injection","authors":"M. Zhao, Hui Zhang, Li Liu, Zhicong Liang, Guang Deng","doi":"10.1109/IPTA.2018.8608158","DOIUrl":"https://doi.org/10.1109/IPTA.2018.8608158","url":null,"abstract":"At present, the detection of pharmaceutical injection products is a quite important step in the pharmaceutical manufacturing, as it has the direct related to the quality of medical product quality. Aiming at the difficulty that liquid particle has a smaller pixel point in the high resolution image of detection of pharmaceutical liquid particle, hence consider combined with deep neural network and clustering algorithm for detection and localization of little particle, and a processing method combining single frame images with multi-frame images was proposed to identifying liquid particle. Firstly, the single-frame image is detected by using Faster-RCNN deep neural network, and it can obtain the detection result of the 8-frame sequence image. Then hierarchical clustering and K-means clustering algorithm are used for clustering to obtain the same target motion area. In this way, liquid particle can be more accurately identified and the accuracy of detection can be greatly improved. The experimental results show that the accuracy of detection and recognition of foreign substances in liquid medicine is improved by more than 10% on average.","PeriodicalId":272294,"journal":{"name":"2018 Eighth International Conference on Image Processing Theory, Tools and Applications (IPTA)","volume":"131 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133721696","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：