Навчання моделей згорткових нейронних мереж виявленню об’єктів, сцен і контекстів на зображеннях

Journal Title: Challenges and Issues of Modern Science - Year 2024, Vol 3, Issue 1

Abstract

Purpose. The study focuses on developing an optimized convolutional neural network (CNN) for detecting objects, scenes, and contexts in diverse images. It emphasizes improving the architecture, training methods, and performance of CNNs in computer vision tasks, which are essential for various industries. Design / Method / Approach. The study uses Python, TensorFlow, and Keras to create and train a CNN on dataset CIFAR-10. Hyperparameter tuning and data augmentation techniques were applied to enhance model performance. Findings. The CNN model trained on CIFAR-10 demonstrated strong performance with an accuracy of approximately 85% on the test set, highlighting its ability to classify diverse objects. Data augmentation techniques significantly improved the overall performance by making the model more robust to image variations. Theoretical Implications. The study emphasizes the importance of proper data preparation, including image normalization and augmentation, to achieve high accuracy in CNN models. Practical Implications. The application of convolutional neural networks in real-world scenarios such as security, medicine, and autonomous systems is transformative. These models can accurately detect objects and understand contexts, opening new possibilities for innovation and automation in various industries. Originality / Value. This study makes a valuable contribution to research on convolutional neural networks by showcasing the successful training and optimization of a CNN for object detection. The combination of data augmentation, architecture design, and hyperparameter tuning highlights an effective approach to achieving high accuracy in computer vision tasks. Research Limitations / Future Research. Future research could explore alternative CNN architectures and larger datasets to further enhance object detection accuracy. Additionally, integrating new learning strategies could improve the model's performance in more complex and varied environments. Paper Type. Applied Research.

Authors and Affiliations

Роман Орлов, Сергій Таборанський

Keywords

Related Articles

Кібербезпека критичної інфраструктури: виклики інновацій і загрози цифрових технологій

Purpose. The article aims to explore the characteristics, development prospects, and potential threats associated with the cybersecurity of critical infrastructure. As critical systems become increasingly dependent on di...

Прогнозування рівня діоксиду сірки в атмосфері від стаціонарних та пересувних джерел забруднення

Purpose. Analysis of the dynamics of changes in the amount of sulfur dioxide entering the atmosphere from stationary and mobile sources of pollution. Design / Method / Approach. The research methodology is based on the u...

Assessment of air pollutants when burning alternative fuels

Purpose. The aim of the work is to identify potential agricultural waste that can become energy sources, which will significantly reduce the cost of the operation of small and medium-sized boilers and provide cost saving...

Методика супутникового моніторингу вирубування лісів за даними Sentinel-2A/B

Purpose. The main goal of the research was development and testing of the methodology of automated processing and analysis of multispectral satellite images of medium spatial resolution for the detection of deforestation...

Digital twin as a tool of the real-time enterprise management mechanism

Purpose. This study is devoted to the actual problem of using digital doubles as a tool for an effective mechanism of real-time enterprise management in the context of modern global challenges. The paper considers the co...

Download PDF file
  • EP ID EP753121
  • DOI https://purl.org/cims/2403.018
  • Views 6
  • Downloads 2

How To Cite

Роман Орлов, Сергій Таборанський (2024). Навчання моделей згорткових нейронних мереж виявленню об’єктів, сцен і контекстів на зображеннях. Challenges and Issues of Modern Science, 3(1), -. https://europub.co.uk./articles/-A-753121