Bimonthly    Since 1986
ISSN 1004-9037
Publication Details
Edited by: Editorial Board of Journal of Data Acquisition and Processing
P.O. Box 2704, Beijing 100190, P.R. China
Sponsored by: Institute of Computing Technology, CAS & China Computer Federation
Undertaken by: Institute of Computing Technology, CAS
Published by: SCIENCE PRESS, BEIJING, CHINA
Distributed by:
China: All Local Post Offices
 
   
      09 May 2023, Volume 38 Issue 3
    Article

    A CUSTOM DEEP CONVOLUTIONAL NEURAL NETWORK CDNN - (WITH YOLO V3 BASED NEWLY CONSTRUCTED BACKBONE) FOR MULTIPLE OBJECT DETECTION
    S.T.Santhanalakshmi1 and Rashmita Khilar 2
    Journal of Data Acquisition and Processing, 2023, 38 (3): 1511-1526 . 

    Abstract

    One of the major fields of research in automation is Object Detection and it has been widely applied in the domains including Image Retrieval (IR), medical diagnosis, Automated Vehicle System (AVS), Surveillance etc. As the application domain increases, the challenges also increased and most of the challenges are still unsolved due to the inefficiency of pre-existing models/architectures in detecting the small and occluded objects in an image. The resolution of input could be increased to detect small objects in a better way, or use techniques such as image sharpening or contrast adjustment to enhance images with low contrast or underexposure. Ensembling can also be done for multiple object detection models to improve overall performance. The proposed research work on CDNN model tries to improve the efficiency in detecting the small and occluded objects without sacrificing the processing time. In the proposed model, the existing backbone has been replaced by custom deep convolutional neural network with added augmentation layers. It is found that the proposed model improved the accuracy of the image detection significantly for small and occluded objects, even if the object is far away from the focus of the camera. With proper feature selection along with hyper parameter tuning, the proposed Custom deep convolutional neural network model (CDNN) resulted with an accuracy rate of 99.02277%.

    Keyword

    Object Detection, Deep Learning, YOLO, Computer Vision, Deep Convolutional Neural Network, Augmentations.


    PDF Download (click here)

SCImago Journal & Country Rank

ISSN 1004-9037

         

Home
Editorial Board
Author Guidelines
Subscription
Journal of Data Acquisition and Processing
Institute of Computing Technology, Chinese Academy of Sciences
P.O. Box 2704, Beijing 100190 P.R. China
E-mail: info@sjcjycl.cn
 
  Copyright ©2015 JCST, All Rights Reserved