An effective structure of multi-modal deep convolutional neural network with adaptive group teaching optimization

• Published in: Soft computing (Berlin, Germany) Vol. 26; no. 15; pp. 7211 - 7232
• Main Authors: Gupta, Vinit, Pawar, Santosh
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.08.2022
• Subjects: Artificial Intelligence; Computational Intelligence; Control; Engineering; Mathematical Logic and Foundations; Mechatronics; Optimization; Robotics; Haar wavelet transform; Feature learning; Adaptive group teaching optimization algorithm; Multi-modal deep convolutional neural network
• ISSN: 1432-7643; 1433-7479
• DOI: 10.1007/s00500-022-07107-7

Deep learning models have been extensively used in pattern recognition and image processing. The conventional deep learning methods highly focused on the learning feature for a unique data type. In our work, a wavelet-based multi-modal deep convolutional neural network with an Adaptive group teaching optimization algorithm is proposed for learning hierarchical features from big data. Initially, the dataset is provided as the input to the network model for learning the optimal features. The wavelet with Haar transform has been implemented within the network to reduce the features by eliminating the redundant data. This step reduces the overall architectural complexities involved. The backpropagation algorithm with the stochastic gradient descent algorithm within the network model layers is utilized to enhance the training step. The adaptive group teaching optimization algorithm is utilized in the proposed network model to update the learned weight values by minimizing the loss function. Thus, the complexity of the architecture and the learning time can be decreased, which will lead to greater accuracy. Two datasets, such as Canadian Institute for advanced research-10 and self-taught learning-10, are used for evaluating the performances of wavelet-based multi-modal deep convolutional neural networks with an Adaptive group teaching optimization algorithm. The simulation results showed the overall performance of the proposed method in terms of accuracy, recall, and precision based on the two datasets is better than the existing methods. The existing methods include convolutional neural network, convolutional autoencoder neural network, and multichannel convolutional neural network.