Pruning deep convolutional neural networks for efficient edge computing in condition assessment of infrastructures

Wu, Rih‐Teng; Singla, Ankush; Jahanshahi, Mohammad R.; Bertino, Elisa; Ko, Bongjun; Verma, Dinesh

doi:10.1111/mice.12449

Cited by 79 publications

(42 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…At present, deep learning methods are popularly applied for various applications. For example, damage identification on structures with images (Cha, Choi, & Büyüköztürk, 2017;Gao, Kong, & Mosalam, 2019;Gao & Mosalam, 2018;Li, Zhao, & Zhou, 2019;Ni, Zhang, & Chen, 2019;Wu et al, 2019;Yang et al, 2018), with sensor measurements (Huang, Beck, & Li, 2019;Y. Zhang, Miyamori, Mikami, & Saito, 2019), concrete property estimation (Rafiei, Khushefati, Demirboga, & Adeli, 2017), and vehicle type detection in real traffic data (Molina-Cabello, Luque-Baena, López-Rubio, & Thurnhofer-Hemsi, 2018).…”

Section: Introductionmentioning

confidence: 99%

Convolutional neural networks for pavement roughness assessment using calibration‐free vehicle dynamics

Jeong

Ditzler

2020

Computer aided Civil Eng

View full text Add to dashboard Cite

Road roughness is a measure of how uncomfortable a ride is, and provides an important indicator for the needs of roadway maintenance or repavement, which is closely tied to the state and federal budget prioritization. As such, accurate and timely monitoring of deteriorating road conditions and following maintenance are essential to improve the overall ride quality on the road. Various technologies, including vehiclemounted laser profiling systems, have been developed and adopted for road roughness (e.g., IRI-International Roughness Index) measurement; however, their high cost limits their use. While recent advances in smartphone technologies allow us to use their embedded accelerometers for road roughness monitoring, the complicated process of necessary vehicle calibration hinders the widespread use of the technology in the actual practices. In this work, a deep learning IRI estimation method is proposed with the goal of using anonymous (i.e., calibration-free) vehicles and their responses measured by smartphones as road roughness sensors. A state-of-the-art deep learning algorithm (i.e., CNN-convolutional neural network) and multimetric vehicle dynamics data (i.e., accelerometer, gyroscope), possibly measured by drivers' smartphones, are employed for the purpose. Optimized CNN architecture and data configuration have been investigated to achieve the best performance. The efficacy of the proposed method has been numerically validated using real road IRI information (i.e., Speedway, Tucson, AZ), real driving speed profiles, and four different types of vehicle data with associated uncertainties.

show abstract

Section: Introductionmentioning

confidence: 99%

Convolutional neural networks for pavement roughness assessment using calibration‐free vehicle dynamics

Jeong

Ditzler

2020

Computer aided Civil Eng

View full text Add to dashboard Cite

show abstract

“…Commercially available portable power‐efficient embedded Artificial Intelligence computing devices such as NVIDIA Jetson TX2 can provide a viable solution to this problem and is a scope for future research. Future studies should also focus on making the network smaller, faster, and consequently more suitable for on‐board real‐time computation by pruning of redundant neurons, which do not contribute significantly to the network outputs, as demonstrated by Wu, et al…”

Section: Resultsmentioning

confidence: 99%

“…The damages were detected by the trained Faster RCNN algorithm with Inception-ResNet-v2 as backbone architecture. Limited computation capability of on-board processing units is another bottleneck in robot-based Future studies should also focus on making the network smaller, faster, and consequently more suitable for on-board real-time computation by pruning of redundant neurons, which do not contribute significantly to the network outputs, as demonstrated by Wu, et al 44…”

Section: Resultsmentioning

confidence: 99%

Deep learning‐based multi‐class damage detection for autonomous post‐disaster reconnaissance

Mondal

Jahanshahi

et al. 2020

Struct Control Health Monit

Self Cite

View full text Add to dashboard Cite

Timely assessment of damages induced to buildings due to an earthquake is critical for ensuring life safety, mitigating financial losses, and expediting the rehabilitation process as well as enhancing the structural resilience where resilience is measured by an infrastructure's capacity to restore full functionality post extreme events. Since manual inspection is expensive, time consuming and risky, low‐cost unmanned aerial vehicles or robots can be leveraged as a viable alternative for quick reconnaissance. Visual data captured by the sensors mounted on the robots can be analyzed, and the damages can be detected and classified autonomously. The present study proposes the use of deep learning‐based approaches to this end. Region‐based convolutional neural network (Faster RCNN) is exploited to detect four different damage types, namely, surface crack, spalling (which includes façade spalling and concrete spalling), and severe damage with exposed rebars and severely buckled rebars. The performance of the proposed approach is evaluated on manually annotated image data collected from reinforced concrete buildings damaged under several past earthquakes such as Nepal (2015), Taiwan (2016), Ecuador (2016), Erzincan (1992), Duzce (1999), Bingol (2003), Peru (2007), Wenchuan (2008), and Haiti (2010). Several experiments are presented in the paper to illustrate the capabilities, as well as the limitations, of the proposed approach for earthquake reconnaissance. It was observed that Inception‐ResNet‐v2 significantly outperforms the other networks considered in this study. The research outcome is a stepping stone forward to facilitate the autonomous assessment of buildings where this can be potentially useful for insurance companies, government agencies, and property owners.

show abstract

“…Deep Learning neural systems, being the latest incarnation of machine learning and currently considered state of the art in the field, have been investigated and applied in different domains such as civil engineering [21][22][23], data clustering [24,25] and many other fields [10,[26][27][28][29][30][31][32][33]. These research activities led to significant improvements also in image segmentation [15,34].…”

Section: Literature Reviewmentioning

confidence: 99%

Shallow buried improvised explosive device detection via convolutional neural networks

Colreavy‐Donnelly

Caraffini

Kühn

et al. 2020

ICA

View full text Add to dashboard Cite

The issue of detecting improvised explosive devices, henceforth IEDs, in rural or built-up urban environments is a persistent and serious concern for governments in the developing world. In many cases, such devices are plastic, or varied metallic objects containing rudimentary explosives, which are not visible to the naked eye and are difficult to detect autonomously. The most effective strategy for detecting land mines also happens to be the most dangerous. This paper intends to leverage the use of a Convolutional Neural Network (CNN) to aid in the discovery of such IEDs. As part of a related project, an autonomous sensor array was used to detect the devices in terrains too hazardous for a human to survey. This paper presents a CNN and its training methodology, suitable to make use of the sensor system. This convolutional neural network can accurately distinguish between a potential IED and surrounding undergrowth and natural features of the environment in real-time. The training methodology enabled the CNN to successfully recognise the IEDs with an accuracy of 98.7%, in well-lit conditions. The results are evaluated against other convolutional neural systems as well as against a deterministic algorithm, showing that the proposed CNN outperforms its competitors including the deterministic method.

show abstract

Pruning deep convolutional neural networks for efficient edge computing in condition assessment of infrastructures

Cited by 79 publications

References 48 publications

Convolutional neural networks for pavement roughness assessment using calibration‐free vehicle dynamics

Convolutional neural networks for pavement roughness assessment using calibration‐free vehicle dynamics

Deep learning‐based multi‐class damage detection for autonomous post‐disaster reconnaissance

Shallow buried improvised explosive device detection via convolutional neural networks

Contact Info

Product

Resources

About