Softmax Deep Boltzmann Cascade Neural Learning Technique For Image Classification

  • G.D.Praveenkumar, Dr. R.Nagaraj
Keywords: Cascade Correlation, Deep Boltzmann Machine, Image Features, Residual Error, Softmax Activation Functions.

Abstract

Classification is a considerable problem to be resolved in image processing.  Recently, few research works have been intended for image classification using various techniques. However, classification accuracy and time complexity of conventional algorithms were not sufficient. In existing techniques, the performance of deep Boltzmann Machine for identifying useful feature representations was lower. In order to overcome such limitations of conventional deep Boltzmann Machine, a novel Softmax Deep Boltzmann Cascade Neural Learning (SDBCNL) Technique is proposed. The SDBCNL Technique developed cascaded layer-wise learning that is applied to Deep Boltzmann architectures for accurate image classification. The SDBCNL Technique reduces the memory and time requirements of the training compared to existing works by learning feature representations that have increased correlation with the output on every layer. The SDBCNL Technique contains three main components namely input unit, hidden unit, and output unit to efficiently carry out the image classification process. The input unit in SDBCNL Technique gets a number of images from CIFAR-10 and CIFAR-100 Image Dataset as input and consequently forwards it to the hidden unit. Then, hidden units in SDBCNL Technique discovers key features such as shape, color, texture, size of objects in each input image. Subsequently, the extracted features at hidden units are forwarded to the output unit. In SDBCNL Technique, the output unit employs the Softmax activation function in order to accurately categorize input image into a multiple class with a lower amount of time consumption. From that, SDBCNL Technique enhances the image classification performance with minimal time. The simulation of the SDBCNL Technique is conducted using metrics such as accuracy, time complexity, false positive rate, and space complexity with respect to a diverse number of images.

Published
2021-09-16
How to Cite
Dr. R.Nagaraj, G. (2021). Softmax Deep Boltzmann Cascade Neural Learning Technique For Image Classification. Design Engineering, 12035- 12053. Retrieved from http://www.thedesignengineering.com/index.php/DE/article/view/4362
Section
Articles