Fine-tuning MobileNetV3 with different weight optimization algorithms for classification of denoised blood cell images using convolutional neural network

A novel method based on convolutional neural networks (CNNs) to denoise the blood cell images (BCI) is proposed in this paper. CNN is a kind of deep learning technique that specializes in retrieving information from input images instantly and capability to reduce the need for expert knowledge when extracting and selecting features. Hyper parameters like activation functions can have a direct impact on the model's performance in CNN. Hence this paper introduced a novel Improved Rectified Linear Units (I-ReLU)-CNNs approach for denoising the BCI images. In addition, the modified-ReLU and NRMSprop are the two techniques used to fine-tune the MobileNetV3 model. Then this fine-tuned MobileNetV3 model is applied for the feature extraction to remove the unwanted features from the original images. Then the Artificial Hummingbird Algorithm (AHA) based on the Manta Ray Foraging optimization algorithm (MRFOA) is proposed for feature selection. Moreover, this AHA-MRFOA is employed to ensure the development of the overall model classification by choosing only the most essential elements. The proposed model is evaluated based on the blood cell image dataset and achieves 97.86% classification accuracy.