Recognition and Classification of Trousers Styles Using Inception_V3 with Contour-Enhanced Preprocessing

Guoding Fu , Hongcai Chen , Hong Liu , Jia Lv , Yan Wang , Zhiyan Huang

Article
2026 / Volume 9 / Pages 1191-1209
Received 28 August 2025; Accepted 15 October 2025; Published 28 April 2026
https://doi.org/10.31881/TLR.2026.1191

Abstract
Trouser style classification on e-commerce platforms suffers from low accuracy and manual inconsistencies, negatively impacting product discovery. This study proposes an automated method using an Inception_V3 convolutional neural network with a systematic preprocessing pipeline combining grayscale conversion, Gaussian blur, gamma correction, and binarisation for high-accuracy classification. A dataset of 4,200 images across seven mainstream trouser styles (straight-leg, wide-leg, bell-bottoms, etc.) was constructed. The strategic preprocessing approach involved converting images to grayscale and subjecting them to contour feature enhancement to amplify silhouette information essential for distinguishing trouser cuts. The pre-trained Inception_V3 model was fine-tuned using transfer learning with Softmax classification. Results demonstrated exceptional efficacy, achieving 96% average classification accuracy on preprocessed images—substantially outperforming unprocessed image training. Specifically, the method achieved macro-average precision of 0.96, recall of 0.96, and F1-score of 0.96 across all seven categories. Performance was consistent across all categories. The 96% accuracy confirms Inception_V3’s effectiveness for apparel classification. Crucially, the performance gap between preprocessed and unprocessed models underscores our contour enhancement strategy’s critical role. By emphasising garment shape over textural details, the model effectively learned to define geometric features, suggesting targeted preprocessing is essential for silhouette-based classification. This research developed an efficient method for automated trouser style classification. The combination of contour-enhancing preprocessing and fine-tuned Inception_V3 provides a robust solution with direct applications for improving inventory management, product tagging, and search functionality on e-commerce platforms.

Keywords
trouser styles, contour enhancement, image preprocessing, deep learning, apparel classification