Fake Clothing Detection Using Deep Learning Method

Abstract

Manufacture and distribution of fake clothing material which can be inferred to be criminal in nature has become a rapidly growing online shopping concern. It can be seen as a way of disguising false information as legitimate one. Indeed, many fashion industries face challenging times to meet market sales and expected profits once fake clothing products are sold on street corners. The consequences of clothing counterfeiting also range from huge losses to buyers and sellers of original products to health hazards, loss of image, and slow growth. More so, while IT has been beneficial, the introduction of IT has also provided a global platform for elusive counterfeiters and traders. The need for efficient/effective techniques for identifying or differentiating original clothing materials from fake ones is consequently on a geometric rise as well. This study developed and evaluated a fake cloth detection model using Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Autoencoder using Python programming. The goal was to improve the capacity for discerning genuine and fake fabric items through image analysis. Dataset acquired from Kaggle was used for the training, testing, and validation phases in the ratio of 70:20:10 respectively. The processes includes resizing the images to a uniform size, converting them to grayscale or applying color normalization, and removing any irrelevant information. Data augmentation methods were applied to enhance the dataset's diversity. Results obtained from the implementation of the model shows that the CNN model achieved perfect precision and accuracy, indicating that it performed well on the dataset. The RNN model achieved 97% precision while the Autoencoder model had a lower precision and accuracy compared to the CNN and RNN models. It correctly identified 63% of the positive instances, but its overall accuracy was 56%, indicating that it struggled with correct classification. These results also highlight the importance of selecting appropriate algorithms that align with the specific task requirements, especially as it found the autoencoder may excel in unsupervised learning scenarios, but its limitations become apparent in supervised classification tasks like fake cloth detection.