A major problem in circuit board remanufacturing is the identification of parametric faults from ... more A major problem in circuit board remanufacturing is the identification of parametric faults from age or stress to the individual passive components. We propose a deep machine learning system for simulating and identifying such faults. A simulated dataset is generated for the most common faults in a circuit. This dataset is used to train deep machine learning classification algorithms to identify and classify the faults. The accuracy of system is measured by comparing with real circuit boards in operation.
This paper explores a new method for representing temporal information found in videos. Dynamic M... more This paper explores a new method for representing temporal information found in videos. Dynamic Mode Decomposition (DMD), a method commonly used to reduce the computational effort for other big-data tasks such as flow calculations, is used in this study to aggregate changes between multiple frames. This is applied to the challenge of human action recognition (HAR) tasks using a Two-Stream architecture. Such an architecture takes two convolutional neural networks (CNNs), one analyzing the spatial data and the other analyzing the temporal data as calculated using DMD. This method is compared against others using two common benchmarks, the UCF-101 dataset and the HMDB-51 dataset, achieving 46.0% accuracy on the UCF-101 and 30.8% on the HMDB-51.
Smoothing filters have been extensively used in image and video analysis. In particular, directio... more Smoothing filters have been extensively used in image and video analysis. In particular, directional smoothers have been employed in motion analysis, edge detection, line parameter estimation, and texture analysis. Such applications often necessitate the use of several directional filters oriented at different angles. However, applying a large number of filters commonly requires a significant amount of computing resources. In such cases, real-time performance may be possibly achieved through utilization of hardware devices having parallel processing capabilities. Additionally, techniques can take advantage of the inherent properties of certain smoothing filters. Such a property is steerability, which implies that the outputs of several filtering operations can be linearly combined in order to produce the output of a directional filter at an arbitrary orientation. Although several efficient FPGA implementations of the convolution operation have been presented in the literature for non-separable and separable, research on steerable filter implementations on FPGA is limited. In this paper, steerable Gaussian smoothers are implemented on an FPGA platform. The technique is compared with a software-based implementation. Performance comparisons indicate that the FPGA technique provides significant speed-up factor of at least ∼6, utilizing only a small percentage of the FPGA resources.
I dedicate this work to the people who have had a profound effect on my life. To my parents, Jogi... more I dedicate this work to the people who have had a profound effect on my life. To my parents, Joginipelly Raj Gopal Rao & J.Prabha Rani; to my uncles, P.Govind Rao and P.Mohan Rao; to my sister and brother, M.Sri Laxmi & J.Anil Kumar and last, but certainly not least my grandparents Potlapally Bapu Rao and P.Bharatamma. The blessings and love you all gave were always with me and encouraged me in stepping forward in life. ii Acknowledgements I would like to thank Dr. Dimitrios Charalampidis, my advisor, for his help, suggestions and guidance throughout the course of my thesis research. I appreciate his direction, supervision of my work and his patience especially for reading, rereading and editing my thesis which helped me in progressing in the right path. I acknowledge my friend Mr. Rajesh Chary for his support throughout my studies without which it would have been impossible for me to get through my Master’s degree. His patience, assistance and insight served as invaluable assets in...
Fish recognition and classification are challenging when performed on video data obtained in non-... more Fish recognition and classification are challenging when performed on video data obtained in non-controlled environments (NCE’s) such as in natural waters. Many NOAA Fisheries surveys use underwater cameras to gather video data for this purpose, which facilitate the analysis of fish populations. Since the amount of data is large, manual data analysis is insufficient. Automatic processing tools are necessary. Most techniques that extract features from fish are in two categories. In the first, features are specific to fish but not necessarily to a particular species. Yet, such measurements are often unreliable when extracted from video obtained in NCE’s, since they strongly depend on the aspect of fish with respect to the camera. In the second, features are generic and may include texture and shape descriptors. Such features do not target specific species of interest. In this paper, we present an automatic technique using Gabor filters to extract characteristic features from two impor...
2020 IEEE International Conference on Big Data (Big Data), 2020
This paper explores a new method for representing temporal information found in videos. Dynamic M... more This paper explores a new method for representing temporal information found in videos. Dynamic Mode Decomposition (DMD), a method commonly used to reduce the computational effort for other big-data tasks such as flow calculations, is used in this study to aggregate changes between multiple frames. This is applied to the challenge of human action recognition (HAR) tasks using a Two-Stream architecture. Such an architecture takes two convolutional neural networks (CNNs), one analyzing the spatial data and the other analyzing the temporal data as calculated using DMD. This method is compared against others using two common benchmarks, the UCF-101 dataset and the HMDB-51 dataset, achieving 46.0% accuracy on the UCF-101 and 30.8% on the HMDB-51.
Proceedings of the 2012 44th Southeastern Symposium on System Theory (SSST), 2012
Smoothing filters have been extensively used in image and video analysis. In particular, directio... more Smoothing filters have been extensively used in image and video analysis. In particular, directional smoothers have been employed in motion analysis, edge detection, line parameter estimation, and texture analysis. Such applications often necessitate the use of several directional filters oriented at different angles. However, applying a large number of filters commonly requires a significant amount of computing resources. In such cases, real-time performance may be possibly achieved through utilization of hardware devices having parallel processing capabilities. Additionally, techniques can take advantage of the inherent properties of certain smoothing filters. Such a property is steerability, which implies that the outputs of several filtering operations can be linearly combined in order to produce the output of a directional filter at an arbitrary orientation. Although several efficient FPGA implementations of the convolution operation have been presented in the literature for non-separable and separable, research on steerable filter implementations on FPGA is limited. In this paper, steerable Gaussian smoothers are implemented on an FPGA platform. The technique is compared with a software-based implementation. Performance comparisons indicate that the FPGA technique provides significant speed-up factor of at least ∼6, utilizing only a small percentage of the FPGA resources.
Proceedings of the 2012 44th Southeastern Symposium on System Theory (SSST), 2012
ABSTRACT Thresholding is an important process in many image processing applications. Recently, a ... more ABSTRACT Thresholding is an important process in many image processing applications. Recently, a bi-level image thresholding method based on graph cut was proposed. The method provided thresholding results which were superior to those obtained with previous techniques. Moreover, the technique was computationally less complex compared to other graph cut-based image thresholding approaches. However, the execution time requirements may still be significant, especially if it is of interest to perform real-time thresholding of a large number of images, such as in the case of high-resolution video sequences. In this paper, we propose a method based on the previously proposed graph cut thresholding method, which is nevertheless appropriate for hardware (FPGA) real-time implementations. A subset of the proposed modifications are also appropriate for a general software implementation. Considering only this subset, the C implementation of the modified method is approximately 2.2 times faster than the original method, as it was presented in the original graph cut-based thresholding paper. Furthermore, the FPGA-based implementation is designed to be 70–100 times faster than the software implementation, depending on the image used.
2020 Pan Pacific Microelectronics Symposium (Pan Pacific), 2020
A major problem in circuit board remanufacturing is the identification of parametric faults from ... more A major problem in circuit board remanufacturing is the identification of parametric faults from age or stress to the individual passive components. We propose a deep machine learning system for simulating and identifying such faults. A simulated dataset is generated for the most common faults in a circuit. This dataset is used to train deep machine learning classification algorithms to identify and classify the faults. The accuracy of system is measured by comparing with real circuit boards in operation.
A major problem in circuit board remanufacturing is the identification of parametric faults from ... more A major problem in circuit board remanufacturing is the identification of parametric faults from age or stress to the individual passive components. We propose a deep machine learning system for simulating and identifying such faults. A simulated dataset is generated for the most common faults in a circuit. This dataset is used to train deep machine learning classification algorithms to identify and classify the faults. The accuracy of system is measured by comparing with real circuit boards in operation.
This paper explores a new method for representing temporal information found in videos. Dynamic M... more This paper explores a new method for representing temporal information found in videos. Dynamic Mode Decomposition (DMD), a method commonly used to reduce the computational effort for other big-data tasks such as flow calculations, is used in this study to aggregate changes between multiple frames. This is applied to the challenge of human action recognition (HAR) tasks using a Two-Stream architecture. Such an architecture takes two convolutional neural networks (CNNs), one analyzing the spatial data and the other analyzing the temporal data as calculated using DMD. This method is compared against others using two common benchmarks, the UCF-101 dataset and the HMDB-51 dataset, achieving 46.0% accuracy on the UCF-101 and 30.8% on the HMDB-51.
Smoothing filters have been extensively used in image and video analysis. In particular, directio... more Smoothing filters have been extensively used in image and video analysis. In particular, directional smoothers have been employed in motion analysis, edge detection, line parameter estimation, and texture analysis. Such applications often necessitate the use of several directional filters oriented at different angles. However, applying a large number of filters commonly requires a significant amount of computing resources. In such cases, real-time performance may be possibly achieved through utilization of hardware devices having parallel processing capabilities. Additionally, techniques can take advantage of the inherent properties of certain smoothing filters. Such a property is steerability, which implies that the outputs of several filtering operations can be linearly combined in order to produce the output of a directional filter at an arbitrary orientation. Although several efficient FPGA implementations of the convolution operation have been presented in the literature for non-separable and separable, research on steerable filter implementations on FPGA is limited. In this paper, steerable Gaussian smoothers are implemented on an FPGA platform. The technique is compared with a software-based implementation. Performance comparisons indicate that the FPGA technique provides significant speed-up factor of at least ∼6, utilizing only a small percentage of the FPGA resources.
I dedicate this work to the people who have had a profound effect on my life. To my parents, Jogi... more I dedicate this work to the people who have had a profound effect on my life. To my parents, Joginipelly Raj Gopal Rao & J.Prabha Rani; to my uncles, P.Govind Rao and P.Mohan Rao; to my sister and brother, M.Sri Laxmi & J.Anil Kumar and last, but certainly not least my grandparents Potlapally Bapu Rao and P.Bharatamma. The blessings and love you all gave were always with me and encouraged me in stepping forward in life. ii Acknowledgements I would like to thank Dr. Dimitrios Charalampidis, my advisor, for his help, suggestions and guidance throughout the course of my thesis research. I appreciate his direction, supervision of my work and his patience especially for reading, rereading and editing my thesis which helped me in progressing in the right path. I acknowledge my friend Mr. Rajesh Chary for his support throughout my studies without which it would have been impossible for me to get through my Master’s degree. His patience, assistance and insight served as invaluable assets in...
Fish recognition and classification are challenging when performed on video data obtained in non-... more Fish recognition and classification are challenging when performed on video data obtained in non-controlled environments (NCE’s) such as in natural waters. Many NOAA Fisheries surveys use underwater cameras to gather video data for this purpose, which facilitate the analysis of fish populations. Since the amount of data is large, manual data analysis is insufficient. Automatic processing tools are necessary. Most techniques that extract features from fish are in two categories. In the first, features are specific to fish but not necessarily to a particular species. Yet, such measurements are often unreliable when extracted from video obtained in NCE’s, since they strongly depend on the aspect of fish with respect to the camera. In the second, features are generic and may include texture and shape descriptors. Such features do not target specific species of interest. In this paper, we present an automatic technique using Gabor filters to extract characteristic features from two impor...
2020 IEEE International Conference on Big Data (Big Data), 2020
This paper explores a new method for representing temporal information found in videos. Dynamic M... more This paper explores a new method for representing temporal information found in videos. Dynamic Mode Decomposition (DMD), a method commonly used to reduce the computational effort for other big-data tasks such as flow calculations, is used in this study to aggregate changes between multiple frames. This is applied to the challenge of human action recognition (HAR) tasks using a Two-Stream architecture. Such an architecture takes two convolutional neural networks (CNNs), one analyzing the spatial data and the other analyzing the temporal data as calculated using DMD. This method is compared against others using two common benchmarks, the UCF-101 dataset and the HMDB-51 dataset, achieving 46.0% accuracy on the UCF-101 and 30.8% on the HMDB-51.
Proceedings of the 2012 44th Southeastern Symposium on System Theory (SSST), 2012
Smoothing filters have been extensively used in image and video analysis. In particular, directio... more Smoothing filters have been extensively used in image and video analysis. In particular, directional smoothers have been employed in motion analysis, edge detection, line parameter estimation, and texture analysis. Such applications often necessitate the use of several directional filters oriented at different angles. However, applying a large number of filters commonly requires a significant amount of computing resources. In such cases, real-time performance may be possibly achieved through utilization of hardware devices having parallel processing capabilities. Additionally, techniques can take advantage of the inherent properties of certain smoothing filters. Such a property is steerability, which implies that the outputs of several filtering operations can be linearly combined in order to produce the output of a directional filter at an arbitrary orientation. Although several efficient FPGA implementations of the convolution operation have been presented in the literature for non-separable and separable, research on steerable filter implementations on FPGA is limited. In this paper, steerable Gaussian smoothers are implemented on an FPGA platform. The technique is compared with a software-based implementation. Performance comparisons indicate that the FPGA technique provides significant speed-up factor of at least ∼6, utilizing only a small percentage of the FPGA resources.
Proceedings of the 2012 44th Southeastern Symposium on System Theory (SSST), 2012
ABSTRACT Thresholding is an important process in many image processing applications. Recently, a ... more ABSTRACT Thresholding is an important process in many image processing applications. Recently, a bi-level image thresholding method based on graph cut was proposed. The method provided thresholding results which were superior to those obtained with previous techniques. Moreover, the technique was computationally less complex compared to other graph cut-based image thresholding approaches. However, the execution time requirements may still be significant, especially if it is of interest to perform real-time thresholding of a large number of images, such as in the case of high-resolution video sequences. In this paper, we propose a method based on the previously proposed graph cut thresholding method, which is nevertheless appropriate for hardware (FPGA) real-time implementations. A subset of the proposed modifications are also appropriate for a general software implementation. Considering only this subset, the C implementation of the modified method is approximately 2.2 times faster than the original method, as it was presented in the original graph cut-based thresholding paper. Furthermore, the FPGA-based implementation is designed to be 70–100 times faster than the software implementation, depending on the image used.
2020 Pan Pacific Microelectronics Symposium (Pan Pacific), 2020
A major problem in circuit board remanufacturing is the identification of parametric faults from ... more A major problem in circuit board remanufacturing is the identification of parametric faults from age or stress to the individual passive components. We propose a deep machine learning system for simulating and identifying such faults. A simulated dataset is generated for the most common faults in a circuit. This dataset is used to train deep machine learning classification algorithms to identify and classify the faults. The accuracy of system is measured by comparing with real circuit boards in operation.
Uploads
Papers