Pub Date : 2019-10-01DOI: 10.1109/AIPR47015.2019.9174594
Cindy Gonzales, W. Sakla
Earth observation and remote sensing technologies are widely used in various application areas. Because the abundance of collected data requires automated analytics, many communities are utilizing deep convolutional neural networks for such tasks. Automating cloud detection in remote sensing and earth observation imagery is a useful prerequisite for providing quality imagery for further analysis. In this paper, we train a model that uses a deep convolutional U-Net architecture, utilizing transfer learning to perform semantic segmentation of clouds in satellite imagery. Our proposed model outperforms state-of-the-art networks on a benchmark dataset based on several relevant segmentation metrics, including Jaccard Index (+7.69%), precision (+6.21%), and specificity (+0.37%). Moreover, we demonstrate that transfer learning utilizing a 4-channel input into a U-Net architecture is possible and highly performant by using a deep ResNet-style architecture pre-trained on ImageNet for the initialization of weights in three channels (red, green, and blue bands) and random initialization of weights in the fourth channel (near infrared band) of the first convolutional layer of the network.
{"title":"Semantic Segmentation of Clouds in Satellite Imagery Using Deep Pre-trained U-Nets","authors":"Cindy Gonzales, W. Sakla","doi":"10.1109/AIPR47015.2019.9174594","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174594","url":null,"abstract":"Earth observation and remote sensing technologies are widely used in various application areas. Because the abundance of collected data requires automated analytics, many communities are utilizing deep convolutional neural networks for such tasks. Automating cloud detection in remote sensing and earth observation imagery is a useful prerequisite for providing quality imagery for further analysis. In this paper, we train a model that uses a deep convolutional U-Net architecture, utilizing transfer learning to perform semantic segmentation of clouds in satellite imagery. Our proposed model outperforms state-of-the-art networks on a benchmark dataset based on several relevant segmentation metrics, including Jaccard Index (+7.69%), precision (+6.21%), and specificity (+0.37%). Moreover, we demonstrate that transfer learning utilizing a 4-channel input into a U-Net architecture is possible and highly performant by using a deep ResNet-style architecture pre-trained on ImageNet for the initialization of weights in three channels (red, green, and blue bands) and random initialization of weights in the fourth channel (near infrared band) of the first convolutional layer of the network.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134129548","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 : 2019-10-01DOI: 10.1109/AIPR47015.2019.9316541
Ange Lou, Shuyue Guan, Nada Kamona, M. Loew
Breast cancer is the second leading cause of death for women in the U.S. Early detection of breast cancer is key to higher survival rates to breast cancer patients. We are investigating infrared (IR) thermography as a noninvasive adjunct to mammography for breast cancer screening. IR imaging is radiation-free, pain-free, and non-contact. Automatic segmentation of the breast area from the acquired full-size breast IR images will help limit the area for tumor search, as well as reduce the time and effort costs of manual hand segmentation. Autoencoder-like convolutional and deconvolutional neural networks (C-DCNN) had been applied to automatically segment the breast area in IR images in previous studies. In this study, we applied a state-of-the-art deep-learning segmentation model, MultiResUnet, which consists of an encoder part to capture features and a decoder part for precise localization. It was used to segment the breast area by using a set of breast IR images, collected in our clinical trials by imaging breast cancer patients and normal volunteers with our infrared camera (N2 Imager). The database we used has 450 images, acquired from 14 patients and 16 volunteers. We used a thresholding method to remove interference in the raw images and remapped them from the original 16-bit to 8-bit, and then cropped and segmented the 8-bit images manually. Experiments using leave-one-out cross-validation (LOOCV) and comparison with the ground-truth images by using Tanimoto similarity show that the average accuracy of MultiResUnet is 91.47%, which is about 2% higher than that of the autoencoder. MultiResUnet offers a better approach to segment breast IR images than our previous model.
{"title":"Segmentation of Infrared Breast Images Using MultiResUnet Neural Networks","authors":"Ange Lou, Shuyue Guan, Nada Kamona, M. Loew","doi":"10.1109/AIPR47015.2019.9316541","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9316541","url":null,"abstract":"Breast cancer is the second leading cause of death for women in the U.S. Early detection of breast cancer is key to higher survival rates to breast cancer patients. We are investigating infrared (IR) thermography as a noninvasive adjunct to mammography for breast cancer screening. IR imaging is radiation-free, pain-free, and non-contact. Automatic segmentation of the breast area from the acquired full-size breast IR images will help limit the area for tumor search, as well as reduce the time and effort costs of manual hand segmentation. Autoencoder-like convolutional and deconvolutional neural networks (C-DCNN) had been applied to automatically segment the breast area in IR images in previous studies. In this study, we applied a state-of-the-art deep-learning segmentation model, MultiResUnet, which consists of an encoder part to capture features and a decoder part for precise localization. It was used to segment the breast area by using a set of breast IR images, collected in our clinical trials by imaging breast cancer patients and normal volunteers with our infrared camera (N2 Imager). The database we used has 450 images, acquired from 14 patients and 16 volunteers. We used a thresholding method to remove interference in the raw images and remapped them from the original 16-bit to 8-bit, and then cropped and segmented the 8-bit images manually. Experiments using leave-one-out cross-validation (LOOCV) and comparison with the ground-truth images by using Tanimoto similarity show that the average accuracy of MultiResUnet is 91.47%, which is about 2% higher than that of the autoencoder. MultiResUnet offers a better approach to segment breast IR images than our previous model.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"1206 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131617723","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 : 2019-10-01DOI: 10.1109/AIPR47015.2019.9174584
Zheng Chen, Malintha Fernando, Lantao Liu
We propose a simple but effective obstacle- avoiding approach for autonomous robot navigation. The method computes local but safe navigation path and relies only on visual feature information extracted from the environment. To achieve this, we first build a discrete set of candidate navigation points in camera’s field of view; then the obstacle avoiding navigation points are selected by evaluating rewards of all candidate points, where the reward metric consists of point-wise transiting probability, safety consideration, mutual information of features, and feature density. Next, we construct a navigable passage in the free space by generating a series of convex hulls that are adjacent to each other. With the navigable passage constructed, a local path that lies within the passage is planned for the robot to safely navigate through. We evaluate the method in both a real world indoor environment as well as a simulated outdoor environment.
{"title":"A Visual Feature based Obstacle Avoidance Method for Autonomous Navigation","authors":"Zheng Chen, Malintha Fernando, Lantao Liu","doi":"10.1109/AIPR47015.2019.9174584","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174584","url":null,"abstract":"We propose a simple but effective obstacle- avoiding approach for autonomous robot navigation. The method computes local but safe navigation path and relies only on visual feature information extracted from the environment. To achieve this, we first build a discrete set of candidate navigation points in camera’s field of view; then the obstacle avoiding navigation points are selected by evaluating rewards of all candidate points, where the reward metric consists of point-wise transiting probability, safety consideration, mutual information of features, and feature density. Next, we construct a navigable passage in the free space by generating a series of convex hulls that are adjacent to each other. With the navigable passage constructed, a local path that lies within the passage is planned for the robot to safely navigate through. We evaluate the method in both a real world indoor environment as well as a simulated outdoor environment.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"51 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127287848","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 : 2019-10-01DOI: 10.1109/AIPR47015.2019.9174574
Martin Jaszewski, S. Parameswaran
We address the problem of building a blind image denoising network that better adapts to user-defined efficiency and performance requirements. CNN-based architectures such as FFDNet as well as classical methods like BM3D provide fast denoising capability but require the user to specify an approximate noise level. Blind denoising networks like DnCNN and CBDNet are appealing due to their ease of use by non-experts but can be slow. Additionally, these networks are not designed to allow for selecting a reliable operating point based on constraints like available compute, affordable latency, and expected quality. To this end, we propose to develop denoising networks that are tunable to achieve a desired balance between image quality and model size. We seek inspiration from architectures that are tuned for classification, detection, and semantic segmentation on mobile phone CPUs. Incorporating recent advances in architectural building blocks and network architecture search and building upon the success of the DnCNN architectures, we present an efficient convolutional blind image denoising network.
{"title":"Exploring Efficient and Tunable Convolutional Blind Image Denoising Networks","authors":"Martin Jaszewski, S. Parameswaran","doi":"10.1109/AIPR47015.2019.9174574","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174574","url":null,"abstract":"We address the problem of building a blind image denoising network that better adapts to user-defined efficiency and performance requirements. CNN-based architectures such as FFDNet as well as classical methods like BM3D provide fast denoising capability but require the user to specify an approximate noise level. Blind denoising networks like DnCNN and CBDNet are appealing due to their ease of use by non-experts but can be slow. Additionally, these networks are not designed to allow for selecting a reliable operating point based on constraints like available compute, affordable latency, and expected quality. To this end, we propose to develop denoising networks that are tunable to achieve a desired balance between image quality and model size. We seek inspiration from architectures that are tuned for classification, detection, and semantic segmentation on mobile phone CPUs. Incorporating recent advances in architectural building blocks and network architecture search and building upon the success of the DnCNN architectures, we present an efficient convolutional blind image denoising network.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"166 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121626148","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 : 2019-10-01DOI: 10.1109/AIPR47015.2019.9174579
Ashish Kumar, Teena Sharma, N. Verma, P. Sircar, S. Vasikarla
The performance of vision-based algorithms depends on the quality of digital images. The images corrupt with salt and pepper noise during image acquisition and transmission deteriorate the performance of these algorithms. This generates the necessity of enhancement algorithms for noise removal. This paper presents an approach for salt and pepper noise removal from digital images. Firstly, the corrupt pixels are detected using a Gaussian membership function and then denoising of these corrupt pixels is performed by a combination of Gaussian and Guided filter. A digital image contains visual information in the form of pixel intensities and regions with abrupt intensity changes. The image regions with similar pixel intensities are called homogeneous and the regions with abrupt intensity changes are called edges or textures. These regions are responsible for carrying important image details. The objective of image denoising is to retrieve the actual pixel intensities in such regions. The proposed approach aims to identify and denoise the corrupt pixels by salt and pepper noise such that the details present in homogeneous regions and edges remain unchanged. For detection of corrupt pixels, two thresholds are estimated using Gaussian membership function. Then, a combination of Gaussian and Guided filter is used for denoising these detected corrupt pixels. The Gaussian filter helps to assign proper weights to the neighborhood pixel set for averaging. However, the Guided filter helps to maintain the structure of an image at very high noise level. The experiments are performed on standard images used in literature with different noise levels up to 99%. It shows that the proposed approach performs efficiently in terms of peak signal to noise ratio.
{"title":"Detection and Removal of Salt and Pepper Noise by Gaussian Membership Function and Guided Filter","authors":"Ashish Kumar, Teena Sharma, N. Verma, P. Sircar, S. Vasikarla","doi":"10.1109/AIPR47015.2019.9174579","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174579","url":null,"abstract":"The performance of vision-based algorithms depends on the quality of digital images. The images corrupt with salt and pepper noise during image acquisition and transmission deteriorate the performance of these algorithms. This generates the necessity of enhancement algorithms for noise removal. This paper presents an approach for salt and pepper noise removal from digital images. Firstly, the corrupt pixels are detected using a Gaussian membership function and then denoising of these corrupt pixels is performed by a combination of Gaussian and Guided filter. A digital image contains visual information in the form of pixel intensities and regions with abrupt intensity changes. The image regions with similar pixel intensities are called homogeneous and the regions with abrupt intensity changes are called edges or textures. These regions are responsible for carrying important image details. The objective of image denoising is to retrieve the actual pixel intensities in such regions. The proposed approach aims to identify and denoise the corrupt pixels by salt and pepper noise such that the details present in homogeneous regions and edges remain unchanged. For detection of corrupt pixels, two thresholds are estimated using Gaussian membership function. Then, a combination of Gaussian and Guided filter is used for denoising these detected corrupt pixels. The Gaussian filter helps to assign proper weights to the neighborhood pixel set for averaging. However, the Guided filter helps to maintain the structure of an image at very high noise level. The experiments are performed on standard images used in literature with different noise levels up to 99%. It shows that the proposed approach performs efficiently in terms of peak signal to noise ratio.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121645883","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 : 2019-10-01DOI: 10.1109/AIPR47015.2019.9174590
Rumana Aktar, K. Palaniappan, J. Uhlmann
In this paper we examine and apply a linear-time matrix transformation for online interpolation of missing data blocks in frames of a streaming video sequence. We show that the resulting algorithm produces interpolated pixels that are sufficiently consistent within the context of a single frame that the missing block/tile is typically unnoticed by a viewer of the video sequence. Given the strenuous time constraints imposed by streaming video, this is essentially the only standard of performance that can be applied.
{"title":"A Method for Online Interpolation of Packet-Loss Blocks in Streaming Video","authors":"Rumana Aktar, K. Palaniappan, J. Uhlmann","doi":"10.1109/AIPR47015.2019.9174590","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174590","url":null,"abstract":"In this paper we examine and apply a linear-time matrix transformation for online interpolation of missing data blocks in frames of a streaming video sequence. We show that the resulting algorithm produces interpolated pixels that are sufficiently consistent within the context of a single frame that the missing block/tile is typically unnoticed by a viewer of the video sequence. Given the strenuous time constraints imposed by streaming video, this is essentially the only standard of performance that can be applied.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123902610","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 : 2019-10-01DOI: 10.1109/AIPR47015.2019.9174586
O. Layode, Tasmeer Alam, M. Rahman
Skin cancer is one of the most frequent cancers among human beings. Diagnosing an unknown skin lesion is the first step to determine appropriate treatment. This paper proposes an integrated classification and retrieval based Decision Support System (DSS) for skin cancer detection with an `easy to use’ user interface by applying fusion and ensemble techniques in deep feature spaces. The descriptiveness and discriminative power of features extracted from dermoscopic images are critical to achieve good classification and retrieval performances. In this work, several deep features are extracted based on using transfer learning in several pre-trained Convolutional Neural Networks (CNNs) and Logistic Regression and Support Vector Machine (SVM) models are built as ensembles of classifiers on top of these feature vectors. Furthermore, the content-based image retrieval (CBIR) technique uses the same deep features by fusing those in different feature combinations using a canonical correlation analysis. Based on image-based visual queries submitted by dermatologists, this system would respond by displaying relevant images of pigmented skin lesions of past cases as well as classifying the image category as different types of skin cancer. The system has been trained on a dermoscopic image dataset consists of 1300 images of ten different classes. The best classification (85%) and retrieval accuracies are achieved in a test data set when feature fusion and ensemble techniques are used in all available deep feature spaces. This integrated system would reduce the visual observation error of human operators and enhance clinical decision support for early screening of kin cancers.
{"title":"Deep Learning Based Integrated Classification and Image Retrieval System for Early Skin Cancer Detection","authors":"O. Layode, Tasmeer Alam, M. Rahman","doi":"10.1109/AIPR47015.2019.9174586","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174586","url":null,"abstract":"Skin cancer is one of the most frequent cancers among human beings. Diagnosing an unknown skin lesion is the first step to determine appropriate treatment. This paper proposes an integrated classification and retrieval based Decision Support System (DSS) for skin cancer detection with an `easy to use’ user interface by applying fusion and ensemble techniques in deep feature spaces. The descriptiveness and discriminative power of features extracted from dermoscopic images are critical to achieve good classification and retrieval performances. In this work, several deep features are extracted based on using transfer learning in several pre-trained Convolutional Neural Networks (CNNs) and Logistic Regression and Support Vector Machine (SVM) models are built as ensembles of classifiers on top of these feature vectors. Furthermore, the content-based image retrieval (CBIR) technique uses the same deep features by fusing those in different feature combinations using a canonical correlation analysis. Based on image-based visual queries submitted by dermatologists, this system would respond by displaying relevant images of pigmented skin lesions of past cases as well as classifying the image category as different types of skin cancer. The system has been trained on a dermoscopic image dataset consists of 1300 images of ten different classes. The best classification (85%) and retrieval accuracies are achieved in a test data set when feature fusion and ensemble techniques are used in all available deep feature spaces. This integrated system would reduce the visual observation error of human operators and enhance clinical decision support for early screening of kin cancers.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129501461","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 : 2019-10-01DOI: 10.1109/AIPR47015.2019.9174593
A. Groener, Gary Chern, M. D. Pritt
In this work, we compare the detection accuracy and speed of several state-of-the-art models for the task of detecting oil and gas fracking wells and small cars in commercial electrooptical satellite imagery. Several models are studied from the single-stage, two-stage, and multi-stage object detection families of techniques. For the detection of fracking well pads (50m- 250m), we find single-stage detectors provide superior prediction speed while also matching detection performance of their two and multi-stage counterparts. However, for detecting small cars, two-stage and multi-stage models provide substantially higher accuracies at the cost of some speed. We also measure timing results of the sliding window object detection algorithm to provide a baseline for comparison. Some of these models have been incorporated into the Lockheed Martin Globally-Scalable Automated Target Recognition (GATR) framework.
{"title":"A Comparison of Deep Learning Object Detection Models for Satellite Imagery","authors":"A. Groener, Gary Chern, M. D. Pritt","doi":"10.1109/AIPR47015.2019.9174593","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174593","url":null,"abstract":"In this work, we compare the detection accuracy and speed of several state-of-the-art models for the task of detecting oil and gas fracking wells and small cars in commercial electrooptical satellite imagery. Several models are studied from the single-stage, two-stage, and multi-stage object detection families of techniques. For the detection of fracking well pads (50m- 250m), we find single-stage detectors provide superior prediction speed while also matching detection performance of their two and multi-stage counterparts. However, for detecting small cars, two-stage and multi-stage models provide substantially higher accuracies at the cost of some speed. We also measure timing results of the sliding window object detection algorithm to provide a baseline for comparison. Some of these models have been incorporated into the Lockheed Martin Globally-Scalable Automated Target Recognition (GATR) framework.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"327 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132705828","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 : 2019-10-01DOI: 10.1109/AIPR47015.2019.9174571
J. LaRue
Motivated by the neuroscience concept of striatal beat frequency a new method is presented that takes an electroencephalogram of brain-wave activity and searches for spectral components using a sliding bank of windows consisting of individual singleperiod single-frequency sinusoids. This new approach is now practical due to the presence of high performance computing architectures. And note, this is a departure from legacy time-frequency spectrogram approaches which use one sliding window of constant size to calculate all frequency components simultaneously and it differs from the wavelet method because a suite of wavelets are formulated around one base unit and that unit usually includes a convolutional ramp up and ramp down structure. The proposed single-period singlefrequency search method is more tangible in identifying the existence, in time, of a frequency component due to its self-imposed constraint of using a bank of windows consisting of single period sinusoids, the length of each being a function of frequency and sampling rate. The result of this approach is a rendering of the on-off nature of the conceptualized striatal beat frequency components through a visualization of a matrix of correlation coefficients, where the rows are individual frequencies, (presented up to 1000 Hz in this paper), and where the columns indicate the position of each detected striatal beat frequency window of time.
{"title":"Single-Period Single-Frequency (SPSF) Visualization of an EEG’s Striatal Beat Frequency","authors":"J. LaRue","doi":"10.1109/AIPR47015.2019.9174571","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174571","url":null,"abstract":"Motivated by the neuroscience concept of striatal beat frequency a new method is presented that takes an electroencephalogram of brain-wave activity and searches for spectral components using a sliding bank of windows consisting of individual singleperiod single-frequency sinusoids. This new approach is now practical due to the presence of high performance computing architectures. And note, this is a departure from legacy time-frequency spectrogram approaches which use one sliding window of constant size to calculate all frequency components simultaneously and it differs from the wavelet method because a suite of wavelets are formulated around one base unit and that unit usually includes a convolutional ramp up and ramp down structure. The proposed single-period singlefrequency search method is more tangible in identifying the existence, in time, of a frequency component due to its self-imposed constraint of using a bank of windows consisting of single period sinusoids, the length of each being a function of frequency and sampling rate. The result of this approach is a rendering of the on-off nature of the conceptualized striatal beat frequency components through a visualization of a matrix of correlation coefficients, where the rows are individual frequencies, (presented up to 1000 Hz in this paper), and where the columns indicate the position of each detected striatal beat frequency window of time.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"88 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134499579","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 : 2019-10-01DOI: 10.1109/AIPR47015.2019.9174596
Emily E. Berkson, Jared D. VanCor, Steven Esposito, Gary Chern, M. D. Pritt
The low/no-shot problem refers to a lack of available data for training deep learning algorithms. In remote sensing, complete image data sets are rare and do not always include the targets of interest. We propose a method to rapidly generate highfidelity synthetic satellite imagery featuring targets of interest over a range of solar illuminations and platform geometries. Specifically, we used the Digital Imaging and Remote Sensing Image Generation model and a custom image simulator to produce synthetic imagery of C130 aircraft in place of real Worldview-3 imagery. Our synthetic imagery was supplemented with real Worldview-3 images to test the efficacy of training deep learning algorithms with synthetic data. We deliberately chose a challenging test case of distinguishing C130s from other aircraft, or neither. Results show a negligible improvement in automatic target classification when synthetic data is supplemented with a small amount of real imagery. However, training with synthetic data alone only achieves F1-scores in line with a random classifier, suggesting that there is still significant domain mismatch between the real and synthetic datasets.
{"title":"Synthetic Data Generation to Mitigate the Low/No-Shot Problem in Machine Learning","authors":"Emily E. Berkson, Jared D. VanCor, Steven Esposito, Gary Chern, M. D. Pritt","doi":"10.1109/AIPR47015.2019.9174596","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174596","url":null,"abstract":"The low/no-shot problem refers to a lack of available data for training deep learning algorithms. In remote sensing, complete image data sets are rare and do not always include the targets of interest. We propose a method to rapidly generate highfidelity synthetic satellite imagery featuring targets of interest over a range of solar illuminations and platform geometries. Specifically, we used the Digital Imaging and Remote Sensing Image Generation model and a custom image simulator to produce synthetic imagery of C130 aircraft in place of real Worldview-3 imagery. Our synthetic imagery was supplemented with real Worldview-3 images to test the efficacy of training deep learning algorithms with synthetic data. We deliberately chose a challenging test case of distinguishing C130s from other aircraft, or neither. Results show a negligible improvement in automatic target classification when synthetic data is supplemented with a small amount of real imagery. However, training with synthetic data alone only achieves F1-scores in line with a random classifier, suggesting that there is still significant domain mismatch between the real and synthetic datasets.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"16 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129624714","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：