FARE-RNET: Fashion Cloth Retrieval via Egret Swarm Optimization-Based Convolutional Attention Module Integrated ResNet

In e-commerce platforms, fashion image retrieval using visual queries is an important task for retrieving visually similar cloth items. However, existing retrieval systems often suffer from challenges, such as poor edge preservation and irrelevant feature extraction that leads to low accuracy and irrelevant retrieval items. To overcome these limitations, this research proposes a novel FARE-RNET (Fashion Retrieval using Residual Network with Egret Swarm Optimization) framework for efficient and accurate cloth retrieval based on Myntra Fashion Product (MFP) database. The input cloth images are pre-processed by Bilateral Adaptive Filter (BAF) to reduce noise artifacts while preserving edges and texture details. The integrated Residual Networks with Convolutional Attention Module (RN-CAM) is deployed to extract statistical features of median, variance, mean, weighted entropy (WE), and standard deviation (SD). The Egret Swarm optimization (ESO) algorithm is employed for selecting the most relevant features by eliminating the irrelevant features of the images. Dilated Convolutional Neural Networks (D-CNN) for multi-scale contextual retrieval by enabling more accurate identification of visually similar cloth items. The proposed FARE-RNET were evaluated using metrics, such as MSE, MSRE, NMSE, RMSE, and MAPE. The proposed ESO algorithm achieves a lower MSE of 0.248293 by indicating fashion image retrieval task over the existing methods. The SOA, GOA, and DOA models achieve MSE values of 0.256476, 0.268656, and 0.257896, respectively. The proposed FARE-RNET is evaluated at different hyperparameter learning rates 1 and 2, and it outperforms the D-CNN and DOA with MSE of 0.304697 and 0.257995, respectively.