محمدامير اشراقي

White blood cells (WBCs) are a vital component of the human body's immune system, playing a crucial role in protecting against various diseases. Accurate identification and quantification of WBC subpopulations are essential for disease diagnosis and prevention. This research focuses on the automatic identification and classification of WBCs using deep learning-based approaches. Two preprocessing techniques and an ensemble learning approach were employed on the Raabin-WBC dataset, which comprises over 16,000 WBC images. Image augmentation methods based on digital image processing were utilized to enhance and balance the dataset and prevent network overfitting. The ViT transformer-based network and the DenseNet121 convolutional neural network (CNN) were used as feature extraction models. Finally, an ensemble learning approach was implemented to improve the network's performance. Preprocessing of the ViT and DenseNet models led to significant improvements in classification performance. Subsequently, the final model was developed using the ensemble learning approach, achieving an accuracy of 99.31% on the test-A eva‎luation set and 97.87% on the test-B eva‎luation set. The experimental results demonstrate that the performance of deep learning-based feature extraction networks is highly dependent on data preprocessing and augmentation using image processing techniques. Additionally, the integration of results from two feature extraction networks with different architectures plays a significant role in improving overall accuracy. The final results of this research show superior performance compared to previous works in this field.

يادگيري عميق , يادگيري تجمعي , پردازش تصوير ديجيتالي , شناسايي خودكار سلول خوني

deep learning , ensemble learning , automated blood cell detection , digital image processing

Mohammad Amir Eshraghi

Ahmad Ayatollahi