Improvement on Deep Features through Various Enhancement Techniques for Vehicles Classification

In the smart transport network, the classification of vehicles plays a significant role. However, the traditional classification systems of vehicles can not satisfy the specifications of real-time applications because of disparities such as luminescence, weather, noise, and many more factors. Convol...

Full description

Saved in:
Bibliographic Details
Published in:Sensing and imaging 2021-12, Vol.22 (1), Article 41
Main Authors: Kiran, V. Keerthi, Dash, Sonali, Parida, Priyadarsan
Format: Article
Language:English
Subjects:
Artificial neural networks
Automobiles
Classification
Discrete Wavelet Transform
Distribution functions
Electrical Engineering
Engineering
Equalization
Feature extraction
Histograms
Image classification
Imaging
Microwaves
Original Paper
Radiology
RF and Optical Engineering
Support vector machines
Wavelet transforms
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In the smart transport network, the classification of vehicles plays a significant role. However, the traditional classification systems of vehicles can not satisfy the specifications of real-time applications because of disparities such as luminescence, weather, noise, and many more factors. Convolutional neural network (CNN) has gained more attraction since it boosts the recognition performance considerably. CNN comprises various types of pre-trained networks through which features can be extracted. The focus of the paper is to enhance the features of the existing pre-trained networks by integrating with most advanced enhancement techniques for increasing the vehicle recognition rate. This work utilizes different enhancement techniques to achieve improved features. Different CNN networks chosen in this work are Residual networks (ResNet-18, 50, 101), AlexNet, GoogLeNet, DenseNet-201, VGG-19. The enhancement techniques like Discrete Wavelet Transform (DWT), Histogram Equalization (HE), Adaptive gamma correction with a weighting distribution function (AGCWD), Homomorphic Filter (HF), and Joint Histogram Equalization (JHE) are chosen for the suggested method. SoftMax layer of CNN and Support Vector Machine (SVM) are used for the classification task. Extensive experiments are conducted using 1510 images with 10 different classes of Comprehensive Cars-Surveillance view (CompCars) dataset. The results show that the suggested approaches deliver higher recognition rate than the different traditional CNN networks. Among all the suggested integrated models, the best classification result is achieved mostly for the AGCWD integrated with different networks. The suggested approaches outperform many existing methods.
ISSN:1557-2064
1557-2072
DOI:10.1007/s11220-021-00363-1