Visual Quality Assessment of E-Commerce Product Images Using Convolutional Neural Networks

Purpose: High-quality product images are vital in shaping consumer trust and driving engagement on e-commerce platforms. This study proposes a deep learning-based approach for evaluating the visual quality of product images, with the aim of improving the overall customer experience and presentation standards in online marketplaces. Methods: A custom-labeled dataset was developed, containing thousands of product images categorized into five quality levels. A convolutional neural network (CNN) was trained to classify these images based on their visual quality. In addition, two well-known architectures, MobileNetV2 and EfficientNetB0, were trained under identical conditions to serve as benchmarks for performance comparison. Results: The proposed CNN model achieved an accuracy of 94.93%, outperform-ing both MobileNetV2 (76.60%) and EfficientNetB0 (92.77%). It also delivered the highest performance in terms of precision, recall, and F1-score, confirming its effectiveness in this domain. Conclusion: The results highlight the CNN model’s suitability for real-time quality assessment of e-commerce images. Its strong performance and efficiency make it a promising candidate for integration into commercial platforms. Future work will investigate the use of transformer-based models and more diverse training data to further improve accuracy and generalizability.