Search

Poverty is rampant and very crucial issue in developing countries. Therefore, in this paper, we explore the implementation of machine learning on the estimation of poverty by training input data from widely available and accessible open-source, including nighttime lights (NTL) and OpenStreetMap (OSM) data. We propose this approach as a straightforward, cost-effective and alternative option for previous studies which have been done by deep learning. We applied the linear regression and ridge regression algorithm as our baseline models while using random forest regression, gradient boosting regression and xgboost regression to achieve the better performance. We found that our best model can explain approximately 74% of the variation in wealth index from input features of Myanmar. We then created the poverty map in province administrative level for Myanmar, which indicates that conventional machine learning models with open-source data can still be as efficient as deep learning on poverty estimation.

This paper proposes the use of relative barometric pressure sensors for enhancing the accuracy of human activity detection based on a tri-axial accelerometer. In our experiment, we employ a device consisting of a tri-axial accelerometer and a barometric pressure sensor on 12 subjects’ waists while they are performing different sequences of activities of sitting, standing, walking, and lying. We also set another barometric pressure sensor on the wall as a reference for barometric pressure. We compare activity classification with features extracted from the following datasets 1) acceleration data, 2) acceleration and on-body barometric pressure data, and 3) acceleration, on-body barometric pressure, and the reference barometric pressure data. Three classifiers are compared, i.e., K-Nearest Neighbors (KNN), Decision Tree, and Random Forest. The results show that highest classification accuracy can be achieved when acceleration, on-body barometric pressure, and the reference barometric pressure information are used. This dataset provides the highest classification accuracy of 90.4% with the Random Forest algorithm.

The deployment of Wireless Sensor Nodes which often deploy in large number, are required to be self-sustainable by utilizing Energy Harvesters for harvesting ambient energy as the node’s power source. Several energy managements schemes have been proposed, dealing with the intermittent power generation problem introduced by the harvester. However, reliability and transmission consistency of the node are not their primary concern. This study proposes an Energy harvesting powered Wireless Sensor node energy management algorithm which balances between transmitting data with desired interval and providing sensor network reliability by transmitting the nodes status in low harvesting rate. The simulation results show that, with the proposed algorithm the node is able to provide an average 15 second interval data transmission under fluctuating harvesting rate, verified their status periodically and also gives more consistency in data acquisition which is necessary in monitoring application.

In this paper, we propose a method to directly describe the RTL structure of a pipelined RISC- V processor with cache, memory management unit (MMD) and AXI bus interface using C++ language. This processor C++ model serves as a near cycle-accurate simulation model of the RISC- V core, while our C2RTL framework translates the processor C++ model into cycle-accurate RTL description in Verilog-HDL and RTL-equivalent C model. Our design methodology is unique compared to other existing methodologies since both the simulation model and the RTL model are derived from the same C++ source, which greatly simplifies the design verification and optimization processes. The effectiveness of our design methodology is demonstrated on a RISC-V processor which runs Linux OS on an FPGA board as well as significantly short simulation time of the original C++ processor model and RTL-equivalent C model compared to commercial RTL simulator.

Speaker recognition is a task of identifying/verifying individual identity with the help of input voice sample. Speaker recognition is further classified into speaker identification(SI) and speaker verification(SV). Unlike other approaches using ELSDSR speech database, which only focuses on speaker identification performance, proposed work also focuses on speaker verification performance results, along with SI performance. Main goal of this research, is to find out best model for speaker identification and speaker verification system for clean speech database. Comparative study is done between performances by various combinations of features, for speaker identification and speaker verification system with Feedforward Artificial Neural Network(FFANN) and Support Vector Machine(SVM) as an classification technique using ELSDSR voice database. Features named as Linear predictive coding(LPC), Mel frequency cepstral coefficient(MFCC) and Perceptual linear prediction(PLP) are used. All features are tested separately and in fusion among each other’s with FFANN and SVM classifier on MATLAB software. Also Proposed model results are compared with some famous techniques using same ELSDSR database for speaker identification system. By comparing experimental results of proposed model and others model, it is observed that fusion of different features gives better results and speaker identification accuracy increases to 3%-5% when compared with single feature’s result. In addition, proposed method are giving best result of100% accuracy for speaker identification system and 0 equal error rate(EER) value for speaker verification system, when fusion of MFCC, LPC and PLP features are used with ANN and SVM classifier. Therefore, it can be said that fusion of MFCC, LPC and PLP features greatly enhances the speaker recognition system’s performance.

This paper presents a solution for a small area (partial-to-partial) fingerprint verification, which is suitable for mobile devices with small fingerprint sensors. Since images captured from the fingerprint sensor contain very limited minutiae features and less image overlapping areas between the input image and the template image, the minutiae-based method is hardly used. We propose a new approach that consists of two steps:(l) fingerprint alignment using band-limited phase-only correlation (BLPOC) and (2) fingerprint matching using deep convolutional neural network (DCNN). To handle the large displacement between two fingerprints, we added both preprocessing and post-processing steps to make the fingerprint alignment process more robust. The experiment results show that the proposed method achieves better performances compared to other methods.

Gradient Boosted Tree is most effective and standard machine learning algorithm in many fields especially with various type of tabular dataset. Besides, recent industry field and robotics field require high-speed, power efficient and real-time training with enormous data. FPGA is effective device which enable custom domain specific approach to give acceleration as well as power efficiency. We introduce a scalable full hardware implementation of Gradient Boosted Tree training with high performance and flexibility of hyper parameterization. Experimental work shows that our hardware implementation achieved 11–33 times faster than state-of-art GPU acceleration even with small gates and low power FPGA device.

The 2-D Forward-looking Sonar (2D-FLS) is a new technology of sonar signals that increases the performance of underwater exploration of an Autonomous Underwater Vehicle (AUV). Because the AUV is a fully automatic robot that can handle various missions and does not require humans to control its operation, FLS is the best tool for underwater navigation in poor visibility conditions. In this paper, we used an FLS image sequence to estimate the height of the pipeline and create an elevation map containing information about the height in each of the pixels. In our algorithm, we employed the Markov random field to model the elevation map, since elevations of neighboring pixels are more likely to have similar values. Next, the Metropolis Algorithm (MA) is used to find the solution of the maximum a posterori equation.

Autonomous underwater vehicles (AUVs), unmanned underwater vehicles, provide lower operation costs in underwater pipeline inspection than remotely operated vehicles (ROVs), since there is no need for a human operator. However, the navigation algorithm is an important part of the successful deployment of AUVs. In this work, we propose an algorithm that extracts the direction of a pipeline based on the level set method (LSM). This direction shall be used for the automatic navigation of an AUV. The low light and turbid water prohibit the use of an optical camera. Instead, a forward-looking sonar (FLS) was selected to provide input sequence images. Due to speckle noise, the traditional image segmentation algorithm cannot be used. Thus, the LSM is applied to represent the direction of pipelines. Lastly, the post-processing which estimates the direction is used to increase the accuracy. From experiments, our algorithm can extract the pipeline direction with accuracies of as high as 94.33%.

Autonomous underwater vehicles (AUVs), unmanned underwater vehicles, are cheaper to operate than remotely operated vehicles (ROVs), since an AUV does not require a trained human operator to control. As a result, in this paper, we propose the pipeline tracking algorithm using sequences of the forward-looking sonar (FLS) images installed on a preprogram AUV. The FLS is more suitable to be used in the subsurface environment than an optical camera due to the limited visibility and floating particles in the sub-sea surface environment. In our work we first employed the speckle noise reduction to the FLS images. Then, we applied the level-set method (LSM) to extract a pipeline boundary in each frame. Lastly, the post-processing step was designed to link the unconnected segments together to from a complete pipeline. Our algorithm can perform the segmentation and tracking with more than 85% accuracy.

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.

Speaker recognition is a biometric technique which uses individual speakers voice samples as a input for recognition purpose. Main goal of this work is to obtain better accuracy for speaker recognition system for large number of voice database. In this paper, a comparative study is made between various combinations of features for speaker identification system with feedforward artificial neural network(FFANN) and support vector machine for 10 and 20 speakers. Linear predictive coding, Mel frequency cepstral coefficient and zero crossing rate are used as a feature extraction techniques. Each features are tested separately and in combination with FFANN and SVM classifier on Matlab software. For ANN classifier 70% of total database are used for training,15% for validation and remaining 15% data are used for testing, number of hidden layers and number of neurons used are 2 and 80 respectively. It is observed from the result that efficiency of the system is not dropping by increasing number of speakers from 10(320 voice samples) to 20(640 voice samples) when using combination of LPC, MFCC, ZCR features with ANN classifier.

Ultra-wideband (UWB) is a well-known radio technology whose media access control (MAC) protocol, WiMedia MAC, has considerable potential to ensure high-speed and high-quality data communication for wireless personal area networks. However, these benefits involve a heavy computational workload, thereby posing a challenge to the conventional very-large-scale integration (VLSI) approach in terms of providing the required performance and power efficiency. Therefore, this study aims to optimise the VLSI implementation of the WiMedia MAC system by proposing a hybrid shared-memory and message-passing multiprocessor system-on-chip (MPSoC) architecture. The proposed solution combines the state-of-the-art MPSoC technology and application-specific instruction-set processor techniques to (i) accelerate the MAC protocol at task level by using parallel processing, (ii) enable the using of custom instructions to optimise the inter-processor communication by using an explicit message passing mechanism, and (iii) ease the implementation process by using a high-level software/hardware co-design methodology. The proposed platform is implemented on both system-level SystemC for architecture exploration and standard-cell technology for future chip implementation. Experimental results show that the proposed hybrid MPSoC architecture achieves 24% performance improvement and 22% power savings over the conventional shared-memory-only one.

In multiprocessor systems, low-latency synchronization is extremely important to effectively exploit fine-grain data parallelism and improve overall performance. This brief presents an efficient synchronization for embedded distributed multiprocessors. The proposed solution works in a completely decentralized request–response manner via explicit message exchange among the processing elements. Scalable lock and barrier synchronization algorithms, which are derived from the inherent distributed characteristics of the underlying architecture, are proposed to enable fair, orderly, and contention-free synchronization. We implement the proposed synchronization model in a distributed 32-core architecture with a commercial cycle-accurate SystemC simulation platform. Experimental results that show our proposed approach achieves ultralow synchronization latency and almost ideal scalability when the core count scales.

The uncontrollable weather conditions can cause a serious problem to remote sensing imaginary. One of the weather conditions is a resulting from cloud contamination. As a result, this paper proposed the use of the convolutional autoencoder neural networks to remove clouds from cloud-contaminated images by training on a multi-temporal remote sensing dataset. Here, the observations from different spectral bands are assumed to be independent since their spectral responses are usually nonoverlapped. From this assumption, each convolutional autoencoder neural networks are trained with the observation from only one spectral band. In our method, we have three convolutional autoencoder neural networks for red, green and blue spectral bands. The experiments were conducted on both synthesis and real dataset derived from the actual LANDSAT 8 images from the central part of Thailand where our algorithm has shown to have a superb performance.

The Aerial Patrol System using helicopter is an effective method to inspect power transmission line in complicated geographical terrains. However, it uses a large number of man-hour to manually process the video sequence to a useful database. This paper proposes a procedure to automate the analysis using Canny Edge Detection and Hough Transform to detect the object then using Geometry Information to classify the detected object as the power transmission tower. The experimental results from a set of real-world video sequence are shown in term of the receiver operating characteristic curve to verify the algorithm’s detection performance. The result shows that the proposed algorithm performs significantly better than an existing one.

The Aerial Surveillance is an effective method to inspect overhead power transmission line in complicated geographical terrains. However, the practical automatic method is still needed to improve the inspection efficiency. The Canny-Edge detector and Hough transformation are used to extract the power transmission tower straight-line feature. Then, we use object-based image classification to classify the straight-line. The experiments from a set of real-world video sequence have shown the excellent performance of our approach in term of the receiver operating characteristic curve.