An Enhanced Convolutional Neural Network for NoiseResilient Image Classification

While CNNs excel in image classification, their performance deteriorates under noisy conditions. Thispaper introduces an enhanced CNN (ECN) designed to enhance noise resilience while maintaining highaccuracy in image classification tasks. By replacing the ReLU activation function with K-winners andutilizing sparse weight initialization, the ECNK achieves superior performance even in the presence of upto ۴۰% noise. The hybrid ECNK algorithm is also proposed, combining the strengths of CNN with k-nearestneighbors (KNN) to further increase classification accuracy. The model was tested on both the MNISTdataset and the ABIDE dataset for detecting Autism Spectrum Disorder (ASD) from brain MRI scans.Results demonstrate that the ECNK method achieves a classification accuracy of ۹۹.۸% for ASD detection,even under noisy conditions, significantly outperforming traditional CNN methods.

کلیدواژه ها:

Enhanced Convolutional Neural Network-KNN (ECN) ، Noise-Resilient Image Classification ، Autism Spectrum Disorder Detection ، MRI Image Segmentation ، Deep Learning for Medical Imaging ، Kwinners Activation Function ، Multi-Layer Neural Networks