Hamza Djelouat

Compressive sensing based electronic nose platform

Electronic nose (EN) systems play a significant role for gas monitoring and identification in gas plants. Using an EN system which consists of an array of sensors provides a high performance. Nevertheless, this performance is bottlenecked by the high system complexity incorporated with the high number of sensors. In this paper a new EN system is proposed using data sets collected from an in-house fabricated 4×4 tin-oxide gas array sensor. The system exploits the theory of compressive sensing (CS) and distributed compressive sensing (DCS) to reduce the storage capacity and power consumption. The obtained results have shown that compressing the transmitted data to 20% of its original size will preserve the information by achieving a high reconstruction quality. Moreover, exploiting DCS will maintain the same reconstruction quality for just 15% of the original size. This high quality of reconstruction is explored for classification using several classifiers such as decision tree (DT), K-nearest neighbour (KNN) and extended nearest neighbour (ENN) along with linear discrimination analysis (LDA) as feature reduction technique. CS-based reconstructed data has achieved a 95% classification accuracy. Furthermore, DCS-based reconstructed data achieved a 98.33% classification accuracy which is the same as using original data without compression.

An Adaptive Joint Sparsity Recovery for Compressive Sensing Based EEG System

The last decade has witnessed tremendous efforts to shape the internet of thing (IoT) platforms to be well suited for healthcare applications. These applications involve the deployment of remote monitoring platforms to collect different information about several vital signs such as electroencephalogram (EEG). However, the deployment of these platforms faces several limitations in terms of high power consumption and system complexity. High energy consumption associated with the continuous wireless data transmission can be optimized by exploring efficient compression techniques such as compressive sensing (CS). CS is an emerging theory that enables a compressed acquisition using well chosen sensing matrices to take random projections of the data in sub-Nyquist sampling rates. In addition, system complexity can be reduced by using hardware friendly structured sensing matrices. This paper quantifies the performance of CS-based scheme for vital sign acquisition for a connected health application over an IoT platform taking multi-channel EEG signals as a case study. In addition, the paper exploits the joint sparsity of multi-channel EEG signals as well as a designed sparsifying basis to increase the sparsity of the EEG signal to improve the reconstruction quality, hence, increase the efficiency of the system.

Electronic nose system on the Zynq SoC platform

Abstract Electronic nose or machine olfaction are systems used for detection and identification of odorous compounds and gas mixtures. An electronic nose system is mainly made of two parts, the sensing part which takes the form of a single or a set of sensors and the processing part which takes the form of some pattern recognition algorithms. As an alternative solution to pure software or hardware implementation of the processing part of a gas identification system, this paper proposes a hardware/software co-design approach using the Zynq platform for the implementation of an electronic nose system based on principal component analysis as a dimensionality reduction technique and decision tree as a classification algorithm using two different sensors array, a 4 × 4 in-house fabricated sensor and a commercial one based on 7 Figaro sensors, for comparison purpose. The system was successfully trained and simulated in MATLAB environment prior to the implementation on the Zynq platform. Various scenarios were explored and discussed including the investigation of different combination of principal components as well as the utilization of drift compensation technique to improve the identification accuracy. High level synthesis was carried out on the proposed designs using different optimization directives including loop unrolling, array partitioning and pipelining. Hardware implementation results on the Zynq system on chip show that real-time performances can be achieved for proposed electronic nose systems using hardware/software co-design approach with a single ARM processor running at 667 MHz and the programmable logic running at 142 MHz. In addition, using the designed IP cores and for the best scenarios, a gas can be identified in 3.46  μ s using the 4 × 4 sensor and 0.55  μ s using the Figaro sensors. Furthermore, it has been noticed that the choice of the sensor array has an important impact on performances in terms of accuracy and processing time. Finally, it has been demonstrated that the programmable logic of the Zynq platform consumes much less power than the processing system.

Inequality Indexes as Sparsity Measures Applied to Ventricular Ectopic Beats Detection and its Efficient Hardware Implementation

Meeting application requirements under a tight power budget is of a primary importance to enable connected health internet of things applications. This paper considers using sparse representation and well-defined inequality indexes drawn from the theory of inequality to distinguish ventricular ectopic beats (VEBs) from non-VEBs. Our approach involves designing a separate dictionary for each arrhythmia class using a set of labeled training QRS complexes. Sparse representation, based on the designed dictionaries of each new test QRS complex is then calculated. Following this, its class is predicted using the winner-takes-all principle by selecting the class with the highest inequality index. The experiments showed promising results ranging between 80% and 100% for the detection of VEBs considering the patient-specific approach, 80% using cross validation and 70% on unseen data using independent sets for training and testing, respectively. An efficient hardware implementation of the alternating direction method of multipliers algorithm is also presented. The results show that the proposed hardware implementation can classify a QRS complex in

CS-based fall detection for connected health applications

69.3~ms

Design and Performance Evaluation of a Committee Machine for Gas Identification

that use only 0.934 W energy.

Empowering Technology Enabled Care Using IoT and Smart Devices: A Review

Fall-related injuries of elderly people have become a major public-health burden resulting in direct physical, physiological and financial costs to the surfer and indirect societal costs. Automated fall detectors play a central role in reducing these damages and in supporting safety and independency of the seniors. Typically, automated fall detection devices can send real time notifications to the caregivers in case of emergency. In this study, we consider the problem of fall detection of compressively sensed data. The proposed approach involves first, acquiring acceleration data from different subjects using different fall and activities of daily living (ADLs) scenarios by means of shimmer devices. The collected data is then, multiplied by a binary sensing matrix. Two classification approaches were investigated using k-nearest neighbour (KNN) and extended nearest neighbour (ENN), respectively. Our experiments showed promising results with accuracies of up to 91.34 % and 91.73 % on the test set using five and ten folds cross validation respectively.

The Accuracy and Efficacy of Real-Time Compressed ECG Signal Reconstruction on a Heterogeneous Multicore Edge-Device

Selecting the best classifier plays a significant role in the current electronic nose systems that can be deployed for gas applications. For this purpose, this paper presents an empirical study on the performance of three different classifiers, namely, binary decision tree (BDT), K-nearest neighbours (KNN) and extended nearest neighbours (ENN) on gas identification. It has been observed that with BDT and ENN a maximum classification accuracy of up to 96.4\(\%\) and 96.7\(\%\) can be obtained, respectively, whereas in the case of KNN up to 97.0\(\%\) accuracy can be achieved. In addition to the individual classifiers, a committee machine (CM) based on the three classifiers has been designed, with and without feedback mechanism to determine the improvement gained by combining these classifiers. The performance attained by the CM with feedback is 97.44\(\%\) and it is slightly better than the one without feedback, that is 97.2\(\%\).