RESEARCH INTO THE IMPACT OF IMAGE AUGMENTATION ON THE ACCURACY OF NEURAL NETWORK CLASSIFICATION OF TEXTILE WASTE
DOI:
https://doi.org/10.35546/kntu2078-4481.2026.2.35Keywords:
image augmentation, fabric micro-image classification, textile waste, fiber composition, computer vision, MobileNetV2Abstract
The paper investigates the impact of image augmentation on the quality of neural network classification of textile materials by microimages for tasks of automated sorting of textile waste by fiber composition. The initial data were obtained from the open set «Fabric Fiber Composition Micro Image Dataset», formed by the authors and published on the Kaggle platform; the sample contains 756 microimages of three classes («30 – 50», «50 – 70», «70 – 100») with a close to uniform distribution. The experiment was performed using a fixed data division protocol with the allocation of an independent test set of 50% and training and validation parts of 40%/10%. As a base model, a compact classifier based on MobileNetV2 was used with pre-training on ImageNet in the frozen feature extractor and classification head mode, training was carried out for 3 epochs with unchanged hyperparameters. Five types of augmentation (horizontal mirroring, rotation, scaling, contrast change, shift) were compared with the base version without augmentation; the quality was assessed by accuracy and macro-Precision, macro-Recall, macro-F1 on the validation and test sets, as well as by the indicators for each class and the difference ΔF1 relative to the base version. In the test, the best integral result was shown by scaling: macro-F1=0.7968 versus 0.7814 without augmentation (an increase of about 0.015), while in the validation, the maximum macro-F1 was provided by contrast change (0.7842). Class-specific analysis revealed heterogeneity of effects: for the class «30 – 50» scaling gave ΔF1=+0.067, while for the class «70 – 100» rotation and scaling reduced quality (ΔF1=-0.025 and -0.023, respectively). The stability of the conclusions was confirmed by repeated runs: for scaling test macro-F1=0.8047±0.0052. The obtained results justify the feasibility of selecting augmentations taking into account the quality balance between classes and checking reproducibility on an independent test set
References
Faghih E., Saki Z., Moore M. A systematic literature review-AI-enabled textile waste sorting. Sustainability. 2025. Vol. 17, No. 10. P. 4264. DOI: https://doi.org/10.3390/su17104264.
Mazurets O., Zalutska O., Molchanova M., Sobko O., Bukhantsova L., Zakharkevich O. Deep Learning-Driven Fabric Classification: Distinguishing Natural and Synthetic Materials. CEUR Workshop Proceedings. 2025. Vol. 4164. P. 39–51.
Meng S., Pan R., Gao W., Zhou J., Wang J., He W. A multi-task and multi-scale convolutional neural network for automatic recognition of woven fabric pattern. Journal of Intelligent Manufacturing. 2021. Vol. 32, No. 4. P. 1147–1161. DOI: https://doi.org/10.1007/s10845-020-01607-9.
Iqbal Hussain M. A., Khan B., Wang Z., Ding S. Woven fabric pattern recognition and classification based on deep convolutional neural networks. Electronics. 2020. Vol. 9, No. 6. P. 1048. DOI: https://doi.org/10.3390/electronics9061048.
Tan L., Fu Q., Li J. An improved neural network model based on DenseNet for fabric texture recognition. Sensors. 2024. Vol. 24, No. 23. P. 7758. DOI: https://doi.org/10.3390/s24237758.
Lu S., Deng N., Xin B., Wang Y., Wang W. Investigation on image-based digital method for identification on polyester/cotton fiber category. Fibers and Polymers. 2021. Vol. 22, No. 6. P. 1774–1783. DOI: https://doi.org/10.1007/s12221-021-0802-7.
Sobko O., Mazurets O., Molchanova M., Krak I., Barmak O. Method for analysis and formation of representative text datasets. CEUR Workshop Proceedings. 2025. Vol. 3899. P. 84–98.
Molchanova M. O., Didur V. O., Mazurets O. V. Approach to Data Dimensionality Reduction and Defect Classification Based on Vibration Analysis for Maintenance of Rotating Machinery. Radio Electronics, Computer Science, Control. 2025 (1). P. 84–95. DOI: https://doi.org/10.15588/1607-3274-2025-1-8.
Kumar T., Brennan R., Mileo A., Bendechache M. Image data augmentation approaches: A comprehensive survey and future directions. IEEE Access. 2024. Vol. 12. P. 187536–187571. DOI: https://doi.org/10.1109/ACCESS.2024.3470122.
Molchanova M., Didur V., Mazurets O., Sobko O., Zakharkevich O. Method for Construction and Demolition Waste Classification Using Two-Factor Neural Network Image Analysis. CEUR Workshop Proceedings. 2025. Vol. 3970. P. 168–182.
Мазурець О. В., Жарновський О. В., Гладун О. В., Собко О. В. Нейромережеве виявлення фейкових зображень людей. Вісник Хмельницького національного університету. Серія: Технічні науки. 2025. № 5, Т. 1. С. 416–422.
Мазур Є. В., Мазурець О. В., Кліменко В. І., Собко О. В., Залуцька О. О. Алгоритми та програмна архітектура інформаційної системи нейромережевого аналізу постави людини. Вісник Хмельницького національного університету. Серія: Технічні науки. 2025. № 3, Т. 1. С. 275–284.
Молчанова М. О., Мазурець О. В., Залуцька О. О., Кадинська В. Д., Масловська В. В. Інформаційна хмарна технологія нейромережевого аналізу зруйнованих споруд за візуальними даними з БПЛА. Вісник Херсонського національного технічного університету. 2025. № 3 (94), Т. 2. С. 345–351. DOI: https://doi.org/10.35546/kntu2078-4481.2025.3.2.44.
Xing W., Liu Y., Xin B., Zang L., Deng N. The application of deep and transfer learning for identifying cashmere and wool fibers. Journal of Natural Fibers. 2022. Vol. 19, No. 1. P. 88–104. DOI: https://doi.org/10.1080/15440478.2020.1727817.
Bai J., Wu D., Shelley T., Schubel P., Twine D., Russell J., Zhang J. A comprehensive survey on machine learning driven material defect detection. ACM Computing Surveys. 2025. Vol. 57, No. 11. P. 1–36. DOI: https://doi.org/10.1145/3730576.
