Akademik Veri Yönetim Sistemi
4th International Conference on Artificial Intelligence and Applied Mathematics in Engineering, Mohiuddin Ahmed,Paul Haskell-Dowland, Editör, Springer Nature, Chur, ss.135-150, 2023
Python is an interpreted high-level programming language and is especially popular in realizing machine learning and data science applications since it provides various conveniences and rich libraries for developers. TensorFlow is a well-known open-source machine learning library with many utilities to develop machine learning applications. The TensorFlow framework accelerates computation processes due to multicore computing units such as CPUs and GPUs and various compiled algorithms implementing machine learning and deep learning applications. TensorFlow operators defined for tensors usually work in the GPU device and thus perform the computations faster than the sequential implementations. In this study, the Local Directional Pattern (LDirP) algorithm, a feature extraction method in computer vision, was developed with the TensorFlow library. Therefore, the new LDirP algorithm written using the TensorFlow operators can benefit from the GPU hardware acceleration without low-level parallel programming. The new algorithm written using the TensorFlow operators can benefit from the GPU hardware acceleration without low-level parallel programming. The proposed implementation was evaluated using various sizes of images, and the speed-up performance of the TensorFlow-based algorithm was presented using comparative evaluations. The results show that implementing the LDirP method using TensorFlow provides significant acceleration ratios over the naive Python equivalent of the algorithm.