Empirical Research on the Influence of Image Enhancement Technology on Image Classification Performance Based on Deep Learning
Main Article Content
Keywords
image enhancement, deep learning, image classification, ResNet-50, transfer learning
Abstract
To address the problem of insufficient systematic evaluation of image preprocessing technologies, this study adopts the control variable method and quantitative analysis to systematically evaluate the influence of three types of image enhancement technologies (destructive enhancement, conservative enhancement, and combined enhancement) on the performance of an image classification model based on ResNet-50 feature extraction and SVM classification. The CIFAR-10 dataset is used with a sample size of 5,000 images. The experimental results show that destructive enhancement brings an average accuracy decrease of 21.24%, Unsharp Mask in conservative enhancement brings an increase of 3.64%, and Wavelet→Unsharp in combined enhancement brings an increase of 2.26%. The impact of image enhancement on classification performance varies significantly: appropriate enhancement effectively improves model performance, while excessive enhancement impairs it. The findings provide a basis for technology selection in practical applications.
References
- [1] Wu, Chen, & Fu, Y. H. (2025). Research on the application of image preprocessing technology in visual recognition. China Informatization, (4), 65-69.
- [2] Wang, Z. F., Yang, T., Min, H., et al. (2025). Research on the mine fire image recognition method based on deep learning model. China Safety Science and Technology, 21(8), 146-154.
- [3] Chen, C. G. (2024). Research on the supervised contrastive learning algorithm based on mixed image features and regional features [Doctoral dissertation, Henan University]. https://doi.org/10.27114/d.cnki.ghnau.2024.002049.
- [4] Jia, W. B. (2024). Deep learning feature enhancement technology for blurred image classification [Doctoral dissertation, Lanzhou University]. https://doi.org/10.27204/d.cnki.glzhu.2024.001150.
- [5] Liu, P. (2022). Design and implementation of image classification model based on neural network. Wireless Interconnect Technology, 19(22), 53-55.
- [6] Bai, Y. J., Liu, Y., Zhang, P. C., & Gui, Z. G. (2022). Two-stage unsharp masking method based on adaptive gain coefficient. Journal of Test Technology, 36(5), 398-403.
- [7] Liu, T., Sui, X. B., Chen, Q., & Chen, X. L. (2025). Multiscale nonuniformity correction algorithm based on rolling guided filtering in the spatio-temporal domain. Applied Optics, 46(2), 309-318.
- [8] Shi, X. Y., Han, D., Liu, Y. Q., et al. (2025). Research on the application of ResNet50-based image self-recognition technology in urban physical examination. Building Science, 41(10), 249-261. https://doi.org/10.13614/j.cnki.11-1962/tu.2025.10.27.
- [9] Wu, A. Q., & Han, J. (2024). Analysis and suggestions on computer network image recognition system based on machine learning. China Science and Technology Information, (20), 109-111.
Article Sidebar
This work is licensed under a Creative Commons Attribution 4.0 International License.