Multi-Branch Deep Classification Models.

Abstract

Abstract Deep learning classification models, such as Convolutional Neural Networks (CNNs), Residual Networks (ResNets), Inception Networks, and MobileNet, are widely used for image classification tasks. These models employ sophisticated architectures, including specialized layers for feature extraction and hierarchical representation learning. EfficientNet optimizes the trade-off between accuracy and computational efficiency, while Capsule Networks aim to overcome limitations in capturing hierarchical features. Their adaptability, scalability, and effectiveness make them essential tools in various computer vision applications. In this paper, we explored three proposed classification models constructed from two CNNs followed by three branches, each with different configurations of convolution layers, activation functions, and pooling layers. The outputs of these branches were concatenated, and the model proceeded with additional layers, including convolution, activation, fully connected, and softmax layers. The proposed models were tested under three distinct scenarios with varying output configurations. Case 1 (two outputs), the models are evaluated for masked/unmasked classification, resulting in two output classes. Case 2 (50 outputs) aims to identify 50 different classes. Case 3 (85 outputs), the models were tested with an extended set of 85 output classes. This comprehensive testing across different scenarios demonstrated the versatility and applicability of the proposed models for diverse image classification tasks.