FashNet: An Improved Hybrid Deep Learning for Multi-Attribute Fashion Classification using Cuttlefish and Quantum Newton Optimization

Fashion attribute editing aims to modify the semantic attributes of a given fashion image while preserving irrelevant regions. Existing fashion classification models struggle to accurately identify multiple clothing attributes due to occlusion, and high-dimensional redundant features. To overcome these challenges, a novel FashNet has been proposed for hybrid deep learning framework for multi-attribute fashion classification. The proposed system integrates four stages: edge detection using morphological dilation operation (EDMDO) for precise cloth segmentation is employed for precise cloth segmentation by effectively isolating the cloth region. The Improved Cuttlefish Optimization (ICO) algorithm is utilized for feature extraction by efficiently capturing discriminative color, texture, and shape attributes from segmented clothing regions. Modified Quantum Newton Optimization (MQNO) algorithm is employed for optimal feature selection, reducing redundant features and enhancing the model’s learning efficiency and accuracy. Probabilistic Recurrent Neural Network (PRNN) is employed to accurately classifying fashion clothing styles based on key attributes such as sleeve, color, length, material, and collar. From the experimental analysis, the proposed FashNet model improves overall accuracy by 27.62%, 1.98%, and 0.48% compared to ViT, Swin Transformer, and Pyramid Vision Transformer respectively. The proposed FashNet model outperformed all others with an AC of 99.009% and a p-value of 0.008, indicating both strong performance and statistical significance.