Image classification on IoT edge devices: profiling and modeling

Magid, Salma Abdel; Petrini, Francesco Maria; Dezfouli, Behnam

doi:10.1007/s10586-019-02971-9

Cited by 24 publications

(6 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For calculation, only three parameters r, A and B should be known (Algorithm 1; see Appendix A.6.1), and only one memory cell N = 1 should be reserved with the capacity MeW 1 = 4 B (a single type occupies 4 bytes). In practice, it is convenient to reserve a one-dimensional array W 1_alg2 [i], where i = 0-784, and recalculate its values with a subsequent increase in p, W 2 (Algorithm 2; see Appendix A.6.2). In this case, the maximum reserved memory is estimated as MeW 1 ≈3 kB.…”

Section: Resultsmentioning

confidence: 99%

“…where r is a positive parameter. By applying (2) to the elements of the weight matrix W1, the following equation is obtained:…”

Section: Network Architecturementioning

confidence: 99%

“…In the age of neural networks and Internet of Things (IoT), the search for new neural network architectures capable of operating on devices with small amounts of memory (10s of kB of RAM) is becoming an urgent agenda [1][2][3]. The constrained devices possess significantly less processing power and memory than a regular smartphone or modern laptop, and usually do not have a user interface [4].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Neural Network for Low-Memory IoT Devices and MNIST Image Recognition Using Kernels Based on Logistic Map

Velichko

2020

Electronics

View full text Add to dashboard Cite

This study presents a neural network which uses filters based on logistic mapping (LogNNet). LogNNet has a feedforward network structure, but possesses the properties of reservoir neural networks. The input weight matrix, set by a recurrent logistic mapping, forms the kernels that transform the input space to the higher-dimensional feature space. The most effective recognition of a handwritten digit from MNIST-10 occurs under chaotic behavior of the logistic map. The correlation of classification accuracy with the value of the Lyapunov exponent was obtained. An advantage of LogNNet implementation on IoT devices is the significant savings in memory used. At the same time, LogNNet has a simple algorithm and performance indicators comparable to those of the best resource-efficient algorithms available at the moment. The presented network architecture uses an array of weights with a total memory size from 1 to 29 kB and achieves a classification accuracy of 80.3–96.3%. Memory is saved due to the processor, which sequentially calculates the required weight coefficients during the network operation using the analytical equation of the logistic mapping. The proposed neural network can be used in implementations of artificial intelligence based on constrained devices with limited memory, which are integral blocks for creating ambient intelligence in modern IoT environments. From a research perspective, LogNNet can contribute to the understanding of the fundamental issues of the influence of chaos on the behavior of reservoir-type neural networks.

show abstract

Section: Resultsmentioning

confidence: 99%

“…where r is a positive parameter. By applying (2) to the elements of the weight matrix W1, the following equation is obtained:…”

Section: Network Architecturementioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Neural Network for Low-Memory IoT Devices and MNIST Image Recognition Using Kernels Based on Logistic Map

Velichko

2020

Electronics

View full text Add to dashboard Cite

show abstract

“…In [12] Abdel Magid et al present us with an extensive study of the feasibility and performance of image classification using IoT devices, exploring the relationships between different factors that may affect energy consumption. This study focuses on non-DNN approaches to image processing but reveals that other alternatives might be also possible, using non-accelerated hardware.…”

Section: B Hardware Accelerated Image Processing In the Edgementioning

confidence: 99%

Architecture for Orchestrating Dynamic DNN-Powered Image Processing Tasks in Edge and Cloud Devices

Gonzalez-Gil

Robles-Enciso

Martı́nez³

et al. 2021

IEEE Access

View full text Add to dashboard Cite

DNN processing on image streams has opened the possibility for new and innovative applications. Some of those would benefit from performing the computation locally, avoiding incurring into latencies due to data travelling to image processing services in the cloud, and thus allowing for faster response times. New devices like the GPU-accelerated NVIDIA Jetson family, such as the Jetson Nano, are capable of running modern DNN image processing models, offering an affordable, powerful and scalable local alternative to cloud processing. Performing local image processing can also benefit security and even GDPR compliance, potentially easing the deployment of this solutions. Not only that, but local image processing can also bring the possibility of applying these techniques in areas with reduced connectivity, where cloud-based solutions are unfeasible . In this work, we propose an architecture for the orchestration of DNN accelerated image processing on IoT devices, based on FogFlow; an orchestration platform capable of leveraging cloud and edge resources. FogFlow is part of the FIWARE initiative and is based in the NGSI family of standards, widely applied in Smart City, Smart Building and Smart Home solutions, making it an easy-to-integrate technology.INDEX TERMS DNN image processing, IoT, orchestration.

show abstract

“…Current technology advancements are going to replace traditional text with images, 3D-images and real time video in next few years [1]. Edge Computing demands the image processing and classification techniques to lower the cost of IoT System communication procedures [2]. Non-Parametric image classification techniques are important in view of multisource data classification [3].…”

Section: Introductionmentioning

confidence: 99%

Binary Image Classification using Parallel Neural Networks

2020

IJATCSE

View full text Add to dashboard Cite

Image Recognition and Classification techniques drives the applied sciences of Image Processing and Computer Vision. Major challenges of image recognition and classification includes time-consumption, inefficient results, learning delay and poor accuracy in classification of images. Recent advancements in cutting-edge technologies address these issues to a greater extent in view of the software, hardware and technology used. Internet of Things (IoT) increases the demand of image classification using enhanced capabilities of Convolution Neural Networks (CNN). Deep Neural Learning capabilities enhances the accuracy rates due to optimized and shared parameters, reduces time consumed in processing due to the application of multiple filters by using pooling layers. Pooling layers outperform the defects of normalization layers. Stochastic Gradient Descent (SGD) optimizer optimizes the objective function by identifying proper smoothing parameters at low convergence rate. The current work performed using Transfer Learning (TL) using Mobilenet, sequential model using SGD optimizer over ReLU and softmax activation functions. Experimental results prove that the proposed method achieves promising classification accuracy for large data of images.

show abstract

Image classification on IoT edge devices: profiling and modeling

Cited by 24 publications

References 21 publications

Neural Network for Low-Memory IoT Devices and MNIST Image Recognition Using Kernels Based on Logistic Map

Neural Network for Low-Memory IoT Devices and MNIST Image Recognition Using Kernels Based on Logistic Map

Architecture for Orchestrating Dynamic DNN-Powered Image Processing Tasks in Edge and Cloud Devices

Binary Image Classification using Parallel Neural Networks

Contact Info

Product

Resources

About