@article{Kasiri_Shafiee_Li_Wong_Eichel_2017, title={Efficient Deep Network Architecture for Vision-Based Vehicle Detection Keyvan Kasiri,}, volume={3}, url={https://openjournals.uwaterloo.ca/index.php/vsl/article/view/181}, DOI={10.15353/vsnl.v3i1.181}, abstractNote={<p>With the progress in intelligent transportation systems in smart<br />cities, vision-based vehicle detection is becoming an important issue<br />in the vision-based surveillance systems. With the advent of<br />the big data era, deep learning methods have been increasingly<br />employed in the detection, classification, and recognition applications<br />due to their performance accuracy, however, there are still<br />major concerns regarding deployment of such methods in embedded<br />applications. This paper offers an efficient process leveraging<br />the idea of evolutionary deep intelligence on a state-of-the-art deep<br />neural network. Using this approach, the deep neural network is<br />evolved towards a highly sparse set of synaptic weights and clusters.<br />Experimental results for the task of vehicle detection demonstrate<br />that the evolved deep neural network can achieve a substantial<br />improvement in architecture efficiency adapting for GPUaccelerated<br />applications without significant sacrifices in detection<br />accuracy. The architectural efficiency of ~4X-fold and ~2X-fold<br />decrease is obtained in synaptic weights and clusters, respectively,<br />while the accuracy of 92.8% (drop of less than 4% compared to the<br />original network model) is achieved. Detection results and network<br />efficiency for the vehicular application are promising, and opens<br />the door to a wider range of applications in deep learning.</p>}, number={1}, journal={Journal of Computational Vision and Imaging Systems}, author={Kasiri, Keyvan and Shafiee, Mohammad Javad and Li, Francis and Wong, Alexander and Eichel, Justin}, year={2017}, month={Oct.} }