关键词:
Coronavirus disease
Discrete wavelet transform
Min-max normalization
Honey badger optimization
Fire hawk optimization
Bi-directional long short-term memory
Capsule network
DEEP
DIAGNOSIS
摘要:
In recent years, coronavirus disease 2019 (COVID-19) has been identified as one of the most infectious diseases spread all over the world. These existing methods faced limitations like higher computational complexity, inappropriate feature learning, overfitting, etc., in COVID-19 disease diagnosis. Thus, the proposed study aims to design a novel hybrid deep learning network for classifying COVID-19 disease from the Chest X-ray images. Initially, the input samples are pre-processed to improve image contrast using Extended Contrast Limited Adaptive Histogram Equalization (ECLAHE) and min -max normalization for normalizing pixel values between 0 and 1. Then, to reduce the computational complexity issues, the most significant features are extracted by using an enhanced discrete wavelet transform method with Grey Level Run Length Matrix (GLRLM), and the feature dimensionality issue is solved by introducing a new chaotic circle map-based honey badger optimization as feature selection (FS) technique. The COVID-19 disease is effectively detected using the selected feature maps, as proposed by a unique attention-based hybrid Bi-LSTM capsule network model. In order to enhance the ability of the proposed model, the hyperparameters are optimally fine-tuned by using a fire hawk optimization algorithm. The simulation is carried out in Python, and the publicly accessible Kaggle dataset is used for experimentation. The simulation analysis shows that the proposed study achieved better detection performance than other existing methods in terms of accuracy (99.3%), precision (99.01%), recall (98.99%), and mean AUC (98.55%). Thus, the proposed hybrid model 's outcomes were efficient with ablation and K-fold analysis in Chest X-ray-based COVID19 detection.