Revolutionizing Image Recognition: Next-Generation CNN Architectures for Handwritten Digits and Objects
Paper i proceeding, 2024

This study addresses the pressing need for computer systems to interpret digital media images with a level of sophistication comparable to human visual perception. By leveraging Convolutional Neural Networks (CNNs), we introduce two innovative architectures tailored to distinct datasets: the MNIST handwritten digit dataset and the Fashion MNIST dataset. Unlike traditional machine learning methods such as Support Vector Machines (SVM) and Random Forests, our customized CNN models remarkably enhance image attribute comprehension and recognition accuracy. Specifically, the model developed for the MNIST dataset achieved an unprecedented accuracy of 98.71% without any bias, while the Fashion MNIST model reached 91.39%, marking significant advancements over conventional algorithms without any bias. This research showcases the superior efficiency of CNNs in processing and understanding digital images. It underscores the potential of deep learning technologies in bridging the gap between computational systems and human-like visual recognition. Through meticulous experimentation and analysis, we illustrate how deep CNNs require less preparatory work than other image-processing algorithms, setting a new benchmark in computer vision.

Handwriting Recognition

Data Science

Visual Object Recognition

Deep Learning

Deep CNN

Performance Analysis

Författare

Md Nurul Absur

City University of New York (CUNY)

Kazi Fahim Ahmad Nasif

Kennesaw State University

Sourya Saha

City University of New York (CUNY)

Sifat Nawrin Nova

Chalmers, Data- och informationsteknik, Datorteknik

IEEE Symposium on Wireless Technology and Applications, ISWTA

23247843 (ISSN) 23247851 (eISSN)

173-178
9798350351354 (ISBN)

8th IEEE Symposium on Wireless Technology and Applications, ISWTA 2024
Kuala Lumpur, Malaysia,

Ämneskategorier (SSIF 2011)

Datavetenskap (datalogi)

Datorseende och robotik (autonoma system)

DOI

10.1109/ISWTA62130.2024.10651815

Mer information

Senast uppdaterat

2024-09-23