STD-net: saree texture detection via deep learning framework for E-commerce applications

In this modern world, people move fast and most people are very busy in their daily scheduled lives. In such a scenario, E-commerce online shopping is a great time-saver. In general, ladies clothing has numerous characteristics that are hard to designate such as texture, shape, color, print, and length. Moreover, accurate extraction of product features is critical in the analysis of fashion images for product search, and texture detection based on the query images remains a more challenging task. To overcome the aforementioned challenges, a novel deep learning-based saree texture detection network (STD-Net) has been proposed for the rapid classification of saree tactile textures based on the user query. The research work is conducted in three phases: (1) Indian Saree Dataset creation and Pre-processing phase (2) Patch generation phase (3) Texture detection of the query image. Initially, the input images are denoised using SCRAB (scalable range-based adaptive bilateral filter). Afterward, a region-based convolutional neural network (RCNN) is used for segmenting the region of interest viz, the field part of a saree into patches with the augmented and annotated dataset of sarees. Finally, the Modified EfficientNet-B3 which is integrated with the squeeze and excitation attention (SEA) module is used to classify the texture of the sarees. The experimental results disclose that the proposed STD-Net attains a better testing accuracy of 99.1% for the texture classification of saree images.