Deep Feature Selection Model Based on Convolutional Neural Network and Binary Marine Predator Algorithm
Noor Muhammed Noori, Omar Saber Qasim · 2022
most image datasets in classification have several features that negatively affect classification accuracy. Dimensions are reduced and features are selected through the use of several metaheuristic algorithms such as the Marine Predators Algorithm (MPA) after they are converted from a connected region to a discrete region. In this study, We first obtain the features from the dataset images using ResNet-18, a powerful convolution neural networks (CNN) architecture, pretrained on ImageNet. Then the features are taken from fully connected layers (FC 1000) in ResNet-18 and input into the binary marine predator (BMPA) algorithm, where the inspired algorithm identifies the most pertinent features and removes the less pertinent and noisy features to improve the classification process, then the features are input into the support vector machine (SVM) classifier. The proposed approach has high efficiency and the ability to obtain higher classification accuracy and few advantages.