In this paper, we propose a deep convolutional neural network for camera based wildfire detection. We train the via transfer learning and use window analysis strategy to increase fire detection rate. To achieve computational efficiency, calculate frequency response of kernels in dense layers eliminate those filters with low energy impulse response. Moreover, reduce storage edge devices, compare Fourier domain discard similar using cosine similarity measure domain. test performance variety...

We propose a co-design approach for compute-in-memory inference deep neural networks (DNN). use multiplication-free function approximators based on l <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> norm along with co-adapted processing array and compute flow. Using the approach, we overcame many deficiencies in current art of in-SRAM DNN such as need digital-to-analog converters (DACs) at each operating SRAM row/column, high precision...

In this paper, we propose a novel layer based on fast Walsh-Hadamard transform (WHT) and smooth-thresholding to replace 1 × convolution layers in deep neural networks. the WHT domain, denoise domain coefficients using new non-linearity, smoothed version of well-known soft-thresholding operator. We also introduce family multiplication-free operators from basic 2×2 Hadamard implement 3 depthwise separable layers. Using these two types layers, bottleneck MobileNet-V2 reduce network's number...

Convolution has been the core operation of modern deep neural networks. It is well known that convolutions can be implemented in Fourier Transform domain. In this article, we propose to use binary block Walsh–Hadamard transform (WHT) instead transform. We WHT-based layers replace some regular convolution utilize both one-dimensional (1D) and 2D WHTs article. 1D layers, compute WHT input feature map denoise domain coefficients using a nonlinearity obtained by combining soft-thresholding with...

In this article, we address the problem of estimating location gas leak sources using sparse unreliable spatio-temporal chemical sensor data. We pose task underlying signal and predicting source as an inverse problem. For purpose, develop a novel deep-learning projection-based framework. incorporate traditional projection-onto-convex-sets (POCS) iteration in structure deep model to obtain regularized solution that conforms our prior knowledge concentration distribution. use discrete cosine...

We present a non-Euclidean vector product for artificial neural networks. The operator does not require any multiplications while providing correlation information between two vectors. Ordinary neurons inner of propose class networks with the universal approximation property over space Lebesgue integrable functions based on proposed product. In this new network, "product" real numbers is defined as sum their absolute values, sign determined by numbers. This used to construct in R <sup...

In this article, we describe an ember detection method in infrared (IR) video. Embers, also called firebrands, can act as wildfire super-spreaders. We develop a novel neural network with Walsh-Hadamard Transform (WHT) layer to process the IR The WHT is used temporal dimension of video data model high-frequency activity due movements. insert ResNet-18 and obtained higher accuracy compared standard single slice processing entire block. repeat experiments on ResNet-34, but found that sufficient...

We propose a kernel-PCA-based method to detect anomaly in chemical sensors. use temporal signals produced by sensors form vectors perform the principal component analysis (PCA). estimate kernel-covariance matrix of sensor data and compute eigenvector corresponding largest eigenvalue covariance matrix. The can be detected comparing difference between actual reconstructed from dominant eigenvector. In this letter, we introduce new multiplication-free kernel, which is related <inline-formula...

Chemical and infra-red sensors generate distinct responses under similar conditions because of sensor drift, noise or resolution errors. In this paper, we develop novel machine learning methods for detecting identifying VOC Ammonia vapor from time-series data obtained by uncalibrated chemical infrared sensors. We process signals using deep neural networks (DNN). Three network algorithms are utilized purpose. Additive (termed AddNet) based on a multiplication-devoid operator consequently...

In this article, we present a computationally efficient algorithm to detect Volatile Organic Compounds (VOC) leaking out of components used in chemical processes petrochemical refineries and plants. A VOC plume from damaged component appears as dynamic dark cloud infrared videos. We describe two-stage deep neural network structure, taking advantage both spatial temporal structure the texture regions created by plume. first moving pixels which are darker then their neighboring pixels. extract...

In this paper, we propose a computationally efficient deep learning framework to address the issue of sensitivity drift compensation for chemical sensors. The estimates underlying signal from sensor measurements by means neural network with multiplication-free Hadamard transform based layer. addition, an additive which can be efficiently implemented in real-time on low-cost processors. temporal structure performs only one multiplication per "convolution" operation. Both regular and have...

Markers such as CD13 and CD133 have been used to identify Cancer Stem Cells (CSC) in various tissue images. It is highly likely that CSC nuclei appear brown stained liver We observe there a high correlation between the ratio of blue colored images dark H&E Therefore, we recommend pathologist observing many an image may also order staining estimate ratio. In this paper, describe computer vision method based on neural network estimating image. The structure multiplication free operator using...

Sensor drift is a major problem in chemical sensors that requires addressing for reliable and accurate detection of analytes. In this paper, we develop causal convolutional neural network (CNN) with Discrete Cosine Transform (DCT) layer to estimate the signal. DCT module, apply soft-thresholding nonlinearity transform domain denoise data obtain sparse representation The soft-threshold values are learned during training. Our results show layer-based CNNs able produce slowly varying baseline...

We consider a family of vector dot products that can be implemented using sign changes and addition operations only. The are energy-efficient as they avoid the multiplication operation entirely. Moreover, induce <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\ell_{1}$</tex> -norm, thus providing robustness to impulsive noise. First, we analytically prove yield symmetric, positive semi-definite generalized covariance matrices, enabling...

While Volatile Organic Compounds (VOC) and ammonia have a place in our daily lives, their leakage into the environment is harmful to human health. In order prevent detect gaseous leaks of VOCs, cyber-physical system (CPS) comprised ordinary people or first responders proposed. This CPS uses small, low-cost sensors coupled smart phones mobile devices with necessary computation communication capabilities. The efficacy such hinges on its ability address technical challenges stemming from fact...

Power iteration is a fundamental algorithm in data analysis. It extracts the eigenvector corresponding to largest eigenvalue of given matrix. Applications include ranking algorithms, principal component analysis (PCA), among many others. Certain use cases may benefit from alternate, non-linear power methods with low complexity. In this paper, we introduce multiplication-avoiding (MAPI). MAPI replaces standard ℓ <inf xmlns:mml="http://www.w3.org/1998/Math/MathML"...

In this paper, we propose a deep convolutional neural network for camera based wildfire detection. We train the via transfer learning and use window analysis strategy to increase fire detection rate. To achieve computational efficiency, calculate frequency response of kernels in dense layers eliminate those filters with low energy impulse response. Moreover, reduce storage edge devices, compare Fourier domain discard similar using cosine similarity measure domain. test performance variety...

Convolution has been the core operation of modern deep neural networks. It is well-known that convolutions can be implemented in Fourier Transform domain. In this paper, we propose to use binary block Walsh-Hadamard transform (WHT) instead transform. We WHT-based layers replace some regular convolution utilize both one-dimensional (1-D) and two-dimensional (2-D) WHTs paper. 1-D 2-D layers, compute WHT input feature map denoise domain coefficients using a nonlinearity which obtained by...

Electrocardiography (ECG) is a widely used tool for studying and diagnosing the heart diseases. Atrial fibrillation (AF) an irregular often rapid rate that can increase risk of strokes, failure other heart-related complications. In this study, we develop novel effective method to predict potential AF patients using our ECG signal dataset collected in University Illinois Hospital Health Sciences System. We use convolutional neural network (CNN) structure process both signals related health...

We propose a co-design approach for compute-in-memory inference deep neural networks (DNN). use multiplication-free function approximators based on l1 norm along with co-adapted processing array and compute flow. Using the approach, we overcame many deficiencies in current art of in-SRAM DNN processing, such as need DACs at each operating SRAM row/column, high precision ADCs, limited support multi-bit weights, vector-scale parallelism. also an SRAM-immersed successive approximation ADC...

We propose a kernel-PCA based method to detect anomaly in chemical sensors. use temporal signals produced by sensors form vectors perform the Principal Component Analysis (PCA). estimate kernel-covariance matrix of sensor data and compute eigenvector corresponding largest eigenvalue covariance matrix. The can be detected comparing difference between actual reconstructed from dominant eigenvector. In this paper, we introduce new multiplication-free kernel, which is related l1-norm for...

In this work, we present a method for detecting anomalous chemical sensors using contrastive learning-based framework. many practical systems, an array of multiple are used. Some the may malfunction due to sensor drift and poisoning. standard learning, aim is learn representations that will have maximum agreement among data samples same concept while having minimal with from other concepts. adapt learning useful out-of-distribution sample detection. Furthermore, compare proposed framework...